fix(frontend): use bracket notation for tool variable paths

Adds complete German (Deutsch) language support to DocsGPT.

Changes:
- Add de.json with full German translations
- Register German in i18n configuration
- Add German to language selector dropdown

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>

.github/dependabot.yml

@@ -13,7 +13,11 @@ updates:
     directory: "/frontend" # Location of package manifests
     schedule:
       interval: "daily"
+  - package-ecosystem: "npm"
+    directory: "/extensions/react-widget"
+    schedule:
+      interval: "daily"
   - package-ecosystem: "github-actions"
     directory: "/"
     schedule:
-      interval: "daily"
+      interval: "daily"

.github/styles/DocsGPT/Spelling.yml

@@ -0,0 +1,11 @@
+extends: spelling
+level: warning
+message: "Did you really mean '%s'?"
+ignore:
+  - "**/node_modules/**"
+  - "**/dist/**"
+  - "**/build/**"
+  - "**/coverage/**"
+  - "**/public/**"
+  - "**/static/**"
+vocab: DocsGPT

.github/styles/config/vocabularies/DocsGPT/accept.txt

@@ -0,0 +1,46 @@
+Ollama
+Qdrant
+Milvus
+Chatwoot
+Nextra
+VSCode
+npm
+LLMs
+APIs
+Groq
+SGLang
+LMDeploy
+OAuth
+Vite
+LLM
+JSONPath
+UIs
+configs
+uncomment
+qdrant
+vectorstore
+docsgpt
+llm
+GPUs
+kubectl
+Lightsail
+enqueues
+chatbot
+VSCode's
+Shareability
+feedbacks
+automations
+Premade
+Signup
+Repo
+repo
+env
+URl
+agentic
+llama_cpp
+parsable
+SDKs
+boolean
+bool
+hardcode
+EOL

.github/workflows/vale.yml

@@ -0,0 +1,26 @@
+name: Vale Documentation Linter
+
+on:
+  pull_request:
+    paths:
+      - 'docs/**/*.md'
+      - 'docs/**/*.mdx'
+      - '**/*.md'
+      - '.vale.ini'
+      - '.github/styles/**'
+
+jobs:
+  vale:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Vale linter
+        uses: errata-ai/vale-action@v2
+        with:
+          files: docs
+          fail_on_error: false
+          version: 3.0.5
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.gitignore

@@ -2,7 +2,9 @@
 __pycache__/
 *.py[cod]
 *$py.class
+experiments/
 
+experiments
 # C extensions
 *.so
 *.next

.vale.ini

@@ -0,0 +1,5 @@
+MinAlertLevel = warning
+StylesPath = .github/styles
+
+[*.{md,mdx}]
+BasedOnStyles = DocsGPT

CONTRIBUTING.md

@@ -147,5 +147,5 @@ Here's a step-by-step guide on how to contribute to DocsGPT:
 Thank you for considering contributing to DocsGPT! 🙏
 
 ## Questions/collaboration
-Feel free to join our [Discord](https://discord.gg/n5BX8dh8rU). We're very friendly and welcoming to new contributors, so don't hesitate to reach out.
+Feel free to join our [Discord](https://discord.gg/vN7YFfdMpj). We're very friendly and welcoming to new contributors, so don't hesitate to reach out.
 # Thank you so much for considering to contributing DocsGPT!🙏

HACKTOBERFEST.md

@@ -32,7 +32,7 @@ Non-Code Contributions:
 - Before contributing check existing [issues](https://github.com/arc53/DocsGPT/issues) or [create](https://github.com/arc53/DocsGPT/issues/new/choose) an issue and wait to get assigned.
 - Once you are finished with your contribution, please fill in this [form](https://forms.gle/Npaba4n9Epfyx56S8).
 - Refer to the [Documentation](https://docs.docsgpt.cloud/).
-- Feel free to join our [Discord](https://discord.gg/n5BX8dh8rU) server. We're here to help newcomers, so don't hesitate to jump in! Join us [here](https://discord.gg/n5BX8dh8rU).
+- Feel free to join our [Discord](https://discord.gg/vN7YFfdMpj) server. We're here to help newcomers, so don't hesitate to jump in! Join us [here](https://discord.gg/vN7YFfdMpj).
   
 Thank you very much for considering contributing to DocsGPT during Hacktoberfest! 🙏 Your contributions (not just simple typos) could earn you a stylish new t-shirt.
 

README.md

@@ -16,23 +16,16 @@
   <a href="https://github.com/arc53/DocsGPT">![link to main GitHub showing Forks number](https://img.shields.io/github/forks/arc53/docsgpt?style=social)</a>
   <a href="https://github.com/arc53/DocsGPT/blob/main/LICENSE">![link to license file](https://img.shields.io/github/license/arc53/docsgpt)</a>
   <a href="https://www.bestpractices.dev/projects/9907"><img src="https://www.bestpractices.dev/projects/9907/badge"></a>
-  <a href="https://discord.gg/n5BX8dh8rU">![link to discord](https://img.shields.io/discord/1070046503302877216)</a>
+  <a href="https://discord.gg/vN7YFfdMpj">![link to discord](https://img.shields.io/discord/1070046503302877216)</a>
   <a href="https://x.com/docsgptai">![X (formerly Twitter) URL](https://img.shields.io/twitter/follow/docsgptai)</a>
 
-<a href="https://docs.docsgpt.cloud/quickstart">⚡️ Quickstart</a> • <a href="https://app.docsgpt.cloud/">☁️ Cloud Version</a> • <a href="https://discord.gg/n5BX8dh8rU">💬 Discord</a>
+<a href="https://docs.docsgpt.cloud/quickstart">⚡️ Quickstart</a> • <a href="https://app.docsgpt.cloud/">☁️ Cloud Version</a> • <a href="https://discord.gg/vN7YFfdMpj">💬 Discord</a>
 <br>
 <a href="https://docs.docsgpt.cloud/">📖 Documentation</a> • <a href="https://github.com/arc53/DocsGPT/blob/main/CONTRIBUTING.md">👫 Contribute</a> • <a href="https://blog.docsgpt.cloud/">🗞 Blog</a>
 <br>
 
 </div>
 
-<div align="center">
-  <br>
-🎃 <a href="https://github.com/arc53/DocsGPT/blob/main/HACKTOBERFEST.md"> Hacktoberfest Prizes, Rules & Q&A </a> 🎃
-  <br>
-  <br>
-</div>
-
 
 <div align="center">
   <br>

application/agents/agent_creator.py

@@ -1,5 +1,8 @@
 from application.agents.classic_agent import ClassicAgent
 from application.agents.react_agent import ReActAgent
+import logging
+
+logger = logging.getLogger(__name__)
 
 
 class AgentCreator:
@@ -13,4 +16,5 @@ class AgentCreator:
         agent_class = cls.agents.get(type.lower())
         if not agent_class:
             raise ValueError(f"No agent class found for type {type}")
+        
         return agent_class(*args, **kwargs)

application/agents/base.py

@@ -12,7 +12,6 @@ from application.core.settings import settings
 from application.llm.handlers.handler_creator import LLMHandlerCreator
 from application.llm.llm_creator import LLMCreator
 from application.logging import build_stack_data, log_activity, LogContext
-from application.retriever.base import BaseRetriever
 
 logger = logging.getLogger(__name__)
 
@@ -22,11 +21,12 @@ class BaseAgent(ABC):
         self,
         endpoint: str,
         llm_name: str,
-        gpt_model: str,
+        model_id: str,
         api_key: str,
         user_api_key: Optional[str] = None,
         prompt: str = "",
         chat_history: Optional[List[Dict]] = None,
+        retrieved_docs: Optional[List[Dict]] = None,
         decoded_token: Optional[Dict] = None,
         attachments: Optional[List[Dict]] = None,
         json_schema: Optional[Dict] = None,
@@ -34,10 +34,11 @@ class BaseAgent(ABC):
         token_limit: Optional[int] = settings.DEFAULT_AGENT_LIMITS["token_limit"],
         limited_request_mode: Optional[bool] = False,
         request_limit: Optional[int] = settings.DEFAULT_AGENT_LIMITS["request_limit"],
+        compressed_summary: Optional[str] = None,
     ):
         self.endpoint = endpoint
         self.llm_name = llm_name
-        self.gpt_model = gpt_model
+        self.model_id = model_id
         self.api_key = api_key
         self.user_api_key = user_api_key
         self.prompt = prompt
@@ -52,7 +53,9 @@ class BaseAgent(ABC):
             api_key=api_key,
             user_api_key=user_api_key,
             decoded_token=decoded_token,
+            model_id=model_id,
         )
+        self.retrieved_docs = retrieved_docs or []
         self.llm_handler = LLMHandlerCreator.create_handler(
             llm_name if llm_name else "default"
         )
@@ -62,16 +65,19 @@ class BaseAgent(ABC):
         self.token_limit = token_limit
         self.limited_request_mode = limited_request_mode
         self.request_limit = request_limit
+        self.compressed_summary = compressed_summary
+        self.current_token_count = 0
+        self.context_limit_reached = False
 
     @log_activity()
     def gen(
-        self, query: str, retriever: BaseRetriever, log_context: LogContext = None
+        self, query: str, log_context: LogContext = None
     ) -> Generator[Dict, None, None]:
-        yield from self._gen_inner(query, retriever, log_context)
+        yield from self._gen_inner(query, log_context)
 
     @abstractmethod
     def _gen_inner(
-        self, query: str, retriever: BaseRetriever, log_context: LogContext
+        self, query: str, log_context: LogContext
     ) -> Generator[Dict, None, None]:
         pass
 
@@ -150,6 +156,7 @@ class BaseAgent(ABC):
         call_id = getattr(call, "id", None) or str(uuid.uuid4())
 
         # Check if parsing failed
+
         if tool_id is None or action_name is None:
             error_message = f"Error: Failed to parse LLM tool call. Tool name: {getattr(call, 'name', 'unknown')}"
             logger.error(error_message)
@@ -164,13 +171,14 @@ class BaseAgent(ABC):
             yield {"type": "tool_call", "data": {**tool_call_data, "status": "error"}}
             self.tool_calls.append(tool_call_data)
             return "Failed to parse tool call.", call_id
-
         # Check if tool_id exists in available tools
+
         if tool_id not in tools_dict:
             error_message = f"Error: Tool ID '{tool_id}' extracted from LLM call not found in available tools_dict. Available IDs: {list(tools_dict.keys())}"
             logger.error(error_message)
 
             # Return error result
+
             tool_call_data = {
                 "tool_name": "unknown",
                 "call_id": call_id,
@@ -181,7 +189,6 @@ class BaseAgent(ABC):
             yield {"type": "tool_call", "data": {**tool_call_data, "status": "error"}}
             self.tool_calls.append(tool_call_data)
             return f"Tool with ID {tool_id} not found.", call_id
-
         tool_call_data = {
             "tool_name": tools_dict[tool_id]["name"],
             "call_id": call_id,
@@ -223,6 +230,7 @@ class BaseAgent(ABC):
         tm = ToolManager(config={})
 
         # Prepare tool_config and add tool_id for memory tools
+
         if tool_data["name"] == "api_tool":
             tool_config = {
                 "url": tool_data["config"]["actions"][action_name]["url"],
@@ -234,8 +242,8 @@ class BaseAgent(ABC):
             tool_config = tool_data["config"].copy() if tool_data["config"] else {}
             # Add tool_id from MongoDB _id for tools that need instance isolation (like memory tool)
             # Use MongoDB _id if available, otherwise fall back to enumerated tool_id
-            tool_config["tool_id"] = str(tool_data.get("_id", tool_id))
 
+            tool_config["tool_id"] = str(tool_data.get("_id", tool_id))
         tool = tm.load_tool(
             tool_data["name"],
             tool_config=tool_config,
@@ -272,28 +280,83 @@ class BaseAgent(ABC):
             for tool_call in self.tool_calls
         ]
 
+    def _calculate_current_context_tokens(self, messages: List[Dict]) -> int:
+        """
+        Calculate total tokens in current context (messages).
+
+        Args:
+            messages: List of message dicts
+
+        Returns:
+            Total token count
+        """
+        from application.api.answer.services.compression.token_counter import (
+            TokenCounter,
+        )
+
+        return TokenCounter.count_message_tokens(messages)
+
+    def _check_context_limit(self, messages: List[Dict]) -> bool:
+        """
+        Check if we're approaching context limit (80%).
+
+        Args:
+            messages: Current message list
+
+        Returns:
+            True if at or above 80% of context limit
+        """
+        from application.core.model_utils import get_token_limit
+        from application.core.settings import settings
+
+        try:
+            # Calculate current tokens
+            current_tokens = self._calculate_current_context_tokens(messages)
+            self.current_token_count = current_tokens
+
+            # Get context limit for model
+            context_limit = get_token_limit(self.model_id)
+
+            # Calculate threshold (80%)
+            threshold = int(context_limit * settings.COMPRESSION_THRESHOLD_PERCENTAGE)
+
+            # Check if we've reached the limit
+            if current_tokens >= threshold:
+                logger.warning(
+                    f"Context limit approaching: {current_tokens}/{context_limit} tokens "
+                    f"({(current_tokens/context_limit)*100:.1f}%)"
+                )
+                return True
+
+            return False
+
+        except Exception as e:
+            logger.error(f"Error checking context limit: {str(e)}", exc_info=True)
+            return False
+
     def _build_messages(
         self,
         system_prompt: str,
         query: str,
-        retrieved_data: List[Dict],
     ) -> List[Dict]:
-        docs_with_filenames = []
-        for doc in retrieved_data:
-            filename = doc.get("filename") or doc.get("title") or doc.get("source")
-            if filename:
-                chunk_header = str(filename)
-                docs_with_filenames.append(f"{chunk_header}\n{doc['text']}")
-            else:
-                docs_with_filenames.append(doc["text"])
-        docs_together = "\n\n".join(docs_with_filenames)
-        p_chat_combine = system_prompt.replace("{summaries}", docs_together)
-        messages_combine = [{"role": "system", "content": p_chat_combine}]
+        """Build messages using pre-rendered system prompt"""
+        # Append compression summary to system prompt if present
+        if self.compressed_summary:
+            compression_context = (
+                "\n\n---\n\n"
+                "This session is being continued from a previous conversation that "
+                "has been compressed to fit within context limits. "
+                "The conversation is summarized below:\n\n"
+                f"{self.compressed_summary}"
+            )
+            system_prompt = system_prompt + compression_context
+
+        messages = [{"role": "system", "content": system_prompt}]
 
         for i in self.chat_history:
             if "prompt" in i and "response" in i:
-                messages_combine.append({"role": "user", "content": i["prompt"]})
-                messages_combine.append({"role": "assistant", "content": i["response"]})
+                messages.append({"role": "user", "content": i["prompt"]})
+                messages.append({"role": "assistant", "content": i["response"]})
             if "tool_calls" in i:
                 for tool_call in i["tool_calls"]:
                     call_id = tool_call.get("call_id") or str(uuid.uuid4())
@@ -313,29 +376,17 @@ class BaseAgent(ABC):
                         }
                     }
 
-                    messages_combine.append(
+                    messages.append(
                         {"role": "assistant", "content": [function_call_dict]}
                     )
-                    messages_combine.append(
+                    messages.append(
                         {"role": "tool", "content": [function_response_dict]}
                     )
-        messages_combine.append({"role": "user", "content": query})
-        return messages_combine
-
-    def _retriever_search(
-        self,
-        retriever: BaseRetriever,
-        query: str,
-        log_context: Optional[LogContext] = None,
-    ) -> List[Dict]:
-        retrieved_data = retriever.search(query)
-        if log_context:
-            data = build_stack_data(retriever, exclude_attributes=["llm"])
-            log_context.stacks.append({"component": "retriever", "data": data})
-        return retrieved_data
+        messages.append({"role": "user", "content": query})
+        return messages
 
     def _llm_gen(self, messages: List[Dict], log_context: Optional[LogContext] = None):
-        gen_kwargs = {"model": self.gpt_model, "messages": messages}
+        gen_kwargs = {"model": self.model_id, "messages": messages}
 
         if (
             hasattr(self.llm, "_supports_tools")
@@ -343,7 +394,6 @@ class BaseAgent(ABC):
             and self.tools
         ):
             gen_kwargs["tools"] = self.tools
-
         if (
             self.json_schema
             and hasattr(self.llm, "_supports_structured_output")
@@ -357,7 +407,6 @@ class BaseAgent(ABC):
                     gen_kwargs["response_format"] = structured_format
                 elif self.llm_name == "google":
                     gen_kwargs["response_schema"] = structured_format
-
         resp = self.llm.gen_stream(**gen_kwargs)
 
         if log_context:

application/agents/classic_agent.py

@@ -1,32 +1,20 @@
+import logging
 from typing import Dict, Generator
+
 from application.agents.base import BaseAgent
 from application.logging import LogContext
-from application.retriever.base import BaseRetriever
-import logging
 
 logger = logging.getLogger(__name__)
 
 
 class ClassicAgent(BaseAgent):
-    """A simplified agent with clear execution flow.
-
-    Usage:
-    1. Processes a query through retrieval
-    2. Sets up available tools
-    3. Generates responses using LLM
-    4. Handles tool interactions if needed
-    5. Returns standardized outputs
-
-    Easy to extend by overriding specific steps.
-    """
+    """A simplified agent with clear execution flow"""
 
     def _gen_inner(
-        self, query: str, retriever: BaseRetriever, log_context: LogContext
+        self, query: str, log_context: LogContext
     ) -> Generator[Dict, None, None]:
-        # Step 1: Retrieve relevant data
-        retrieved_data = self._retriever_search(retriever, query, log_context)
+        """Core generator function for ClassicAgent execution flow"""
 
-        # Step 2: Prepare tools
         tools_dict = (
             self._get_user_tools(self.user)
             if not self.user_api_key
@@ -34,20 +22,16 @@ class ClassicAgent(BaseAgent):
         )
         self._prepare_tools(tools_dict)
 
-        # Step 3: Build and process messages
-        messages = self._build_messages(self.prompt, query, retrieved_data)
+        messages = self._build_messages(self.prompt, query)
         llm_response = self._llm_gen(messages, log_context)
 
-        # Step 4: Handle the response
         yield from self._handle_response(
             llm_response, tools_dict, messages, log_context
         )
 
-        # Step 5: Return metadata
-        yield {"sources": retrieved_data}
+        yield {"sources": self.retrieved_docs}
         yield {"tool_calls": self._get_truncated_tool_calls()}
 
-        # Log tool calls for debugging
         log_context.stacks.append(
             {"component": "agent", "data": {"tool_calls": self.tool_calls.copy()}}
         )

application/agents/react_agent.py

@@ -1,284 +1,238 @@
-import os
-from typing import Dict, Generator, List, Any
 import logging
+import os
+from typing import Any, Dict, Generator, List
 
 from application.agents.base import BaseAgent
 from application.logging import build_stack_data, LogContext
-from application.retriever.base import BaseRetriever
 
 logger = logging.getLogger(__name__)
 
+MAX_ITERATIONS_REASONING = 10
+
 current_dir = os.path.dirname(
     os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 )
 with open(
     os.path.join(current_dir, "application/prompts", "react_planning_prompt.txt"), "r"
 ) as f:
-    planning_prompt_template = f.read()
+    PLANNING_PROMPT_TEMPLATE = f.read()
 with open(
-    os.path.join(current_dir, "application/prompts", "react_final_prompt.txt"),
-    "r",
+    os.path.join(current_dir, "application/prompts", "react_final_prompt.txt"), "r"
 ) as f:
-    final_prompt_template = f.read()
-
-MAX_ITERATIONS_REASONING = 10
+    FINAL_PROMPT_TEMPLATE = f.read()
 
 
 class ReActAgent(BaseAgent):
+    """
+    Research and Action (ReAct) Agent - Advanced reasoning agent with iterative planning.
+
+    Implements a think-act-observe loop for complex problem-solving:
+    1. Creates a strategic plan based on the query
+    2. Executes tools and gathers observations
+    3. Iteratively refines approach until satisfied
+    4. Synthesizes final answer from all observations
+    """
+
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
         self.plan: str = ""
         self.observations: List[str] = []
 
-    def _extract_content_from_llm_response(self, resp: Any) -> str:
-        """
-        Helper to extract string content from various LLM response types.
-        Handles strings, message objects (OpenAI-like), and streams.
-        Adapt stream handling for your specific LLM client if not OpenAI.
-        """
-        collected_content = []
-        if isinstance(resp, str):
-            collected_content.append(resp)
-        elif (  # OpenAI non-streaming or Anthropic non-streaming (older SDK style)
-            hasattr(resp, "message")
-            and hasattr(resp.message, "content")
-            and resp.message.content is not None
-        ):
-            collected_content.append(resp.message.content)
-        elif (  # OpenAI non-streaming (Pydantic model), Anthropic new SDK non-streaming
-            hasattr(resp, "choices")
-            and resp.choices
-            and hasattr(resp.choices[0], "message")
-            and hasattr(resp.choices[0].message, "content")
-            and resp.choices[0].message.content is not None
-        ):
-            collected_content.append(resp.choices[0].message.content)  # OpenAI
-        elif (  # Anthropic new SDK non-streaming content block
-            hasattr(resp, "content")
-            and isinstance(resp.content, list)
-            and resp.content
-            and hasattr(resp.content[0], "text")
-        ):
-            collected_content.append(resp.content[0].text)  # Anthropic
-        else:
-            # Assume resp is a stream if not a recognized object
-            chunk = None
-            try:
-                for (
-                    chunk
-                ) in (
-                    resp
-                ):  # This will fail if resp is not iterable (e.g. a non-streaming response object)
-                    content_piece = ""
-                    # OpenAI-like stream
-                    if (
-                        hasattr(chunk, "choices")
-                        and len(chunk.choices) > 0
-                        and hasattr(chunk.choices[0], "delta")
-                        and hasattr(chunk.choices[0].delta, "content")
-                        and chunk.choices[0].delta.content is not None
-                    ):
-                        content_piece = chunk.choices[0].delta.content
-                    # Anthropic-like stream (ContentBlockDelta)
-                    elif (
-                        hasattr(chunk, "type")
-                        and chunk.type == "content_block_delta"
-                        and hasattr(chunk, "delta")
-                        and hasattr(chunk.delta, "text")
-                    ):
-                        content_piece = chunk.delta.text
-                    elif isinstance(chunk, str):  # Simplest case: stream of strings
-                        content_piece = chunk
-
-                    if content_piece:
-                        collected_content.append(content_piece)
-            except (
-                TypeError
-            ):  # If resp is not iterable (e.g. a final response object that wasn't caught above)
-                logger.debug(
-                    f"Response type {type(resp)} could not be iterated as a stream. It might be a non-streaming object not handled by specific checks."
-                )
-            except Exception as e:
-                logger.error(
-                    f"Error processing potential stream chunk: {e}, chunk was: {getattr(chunk, '__dict__', chunk) if chunk is not None else 'N/A'}"
-                )
-
-        return "".join(collected_content)
-
     def _gen_inner(
-        self, query: str, retriever: BaseRetriever, log_context: LogContext
+        self, query: str, log_context: LogContext
     ) -> Generator[Dict, None, None]:
-        # Reset state for this generation call
-        self.plan = ""
-        self.observations = []
-        retrieved_data = self._retriever_search(retriever, query, log_context)
+        """Execute ReAct reasoning loop with planning, action, and observation cycles"""
 
-        if self.user_api_key:
-            tools_dict = self._get_tools(self.user_api_key)
-        else:
-            tools_dict = self._get_user_tools(self.user)
+        self._reset_state()
+
+        tools_dict = (
+            self._get_tools(self.user_api_key)
+            if self.user_api_key
+            else self._get_user_tools(self.user)
+        )
         self._prepare_tools(tools_dict)
 
-        docs_together = "\n".join([doc["text"] for doc in retrieved_data])
-        iterating_reasoning = 0
-        while iterating_reasoning < MAX_ITERATIONS_REASONING:
-            iterating_reasoning += 1
-            # 1. Create Plan
-            logger.info("ReActAgent: Creating plan...")
-            plan_stream = self._create_plan(query, docs_together, log_context)
-            current_plan_parts = []
-            yield {"thought": f"Reasoning... (iteration {iterating_reasoning})\n\n"}
-            for line_chunk in plan_stream:
-                current_plan_parts.append(line_chunk)
-                yield {"thought": line_chunk}
-            self.plan = "".join(current_plan_parts)
-            if self.plan:
-                self.observations.append(
-                    f"Plan: {self.plan} Iteration: {iterating_reasoning}"
-                )
+        for iteration in range(1, MAX_ITERATIONS_REASONING + 1):
+            yield {"thought": f"Reasoning... (iteration {iteration})\n\n"}
 
-            max_obs_len = 20000
-            obs_str = "\n".join(self.observations)
-            if len(obs_str) > max_obs_len:
-                obs_str = obs_str[:max_obs_len] + "\n...[observations truncated]"
-            execution_prompt_str = (
-                (self.prompt or "")
-                + f"\n\nFollow this plan:\n{self.plan}"
-                + f"\n\nObservations:\n{obs_str}"
-                + f"\n\nIf there is enough data to complete user query '{query}', Respond with 'SATISFIED' only. Otherwise, continue. Dont Menstion 'SATISFIED' in your response if you are not ready. "
-            )
+            yield from self._planning_phase(query, log_context)
 
-            messages = self._build_messages(execution_prompt_str, query, retrieved_data)
-
-            resp_from_llm_gen = self._llm_gen(messages, log_context)
-
-            initial_llm_thought_content = self._extract_content_from_llm_response(
-                resp_from_llm_gen
-            )
-            if initial_llm_thought_content:
-                self.observations.append(
-                    f"Initial thought/response: {initial_llm_thought_content}"
-                )
-            else:
-                logger.info(
-                    "ReActAgent: Initial LLM response (before handler) had no textual content (might be only tool calls)."
-                )
-            resp_after_handler = self._llm_handler(
-                resp_from_llm_gen, tools_dict, messages, log_context
-            )
-
-            for (
-                tool_call_info
-            ) in (
-                self.tool_calls
-            ):  # Iterate over self.tool_calls populated by _llm_handler
-                observation_string = (
-                    f"Executed Action: Tool '{tool_call_info.get('tool_name', 'N/A')}' "
-                    f"with arguments '{tool_call_info.get('arguments', '{}')}'. Result: '{str(tool_call_info.get('result', ''))[:200]}...'"
-                )
-                self.observations.append(observation_string)
-
-            content_after_handler = self._extract_content_from_llm_response(
-                resp_after_handler
-            )
-            if content_after_handler:
-                self.observations.append(
-                    f"Response after tool execution: {content_after_handler}"
-                )
-            else:
-                logger.info(
-                    "ReActAgent: LLM response after handler had no textual content."
-                )
-
-            if log_context:
-                log_context.stacks.append(
-                    {
-                        "component": "agent_tool_calls",
-                        "data": {"tool_calls": self.tool_calls.copy()},
-                    }
-                )
-
-            yield {"sources": retrieved_data}
-
-            display_tool_calls = []
-            for tc in self.tool_calls:
-                cleaned_tc = tc.copy()
-                if len(str(cleaned_tc.get("result", ""))) > 50:
-                    cleaned_tc["result"] = str(cleaned_tc["result"])[:50] + "..."
-                display_tool_calls.append(cleaned_tc)
-            if display_tool_calls:
-                yield {"tool_calls": display_tool_calls}
-
-            if "SATISFIED" in content_after_handler:
-                logger.info(
-                    "ReActAgent: LLM satisfied with the plan and data. Stopping reasoning."
+            if not self.plan:
+                logger.warning(
+                    f"ReActAgent: No plan generated in iteration {iteration}"
                 )
                 break
+            self.observations.append(f"Plan (iteration {iteration}): {self.plan}")
 
-        # 3. Create Final Answer based on all observations
-        final_answer_stream = self._create_final_answer(
-            query, self.observations, log_context
-        )
-        for answer_chunk in final_answer_stream:
-            yield {"answer": answer_chunk}
-        logger.info("ReActAgent: Finished generating final answer.")
+            satisfied = yield from self._execution_phase(query, tools_dict, log_context)
 
-    def _create_plan(
-        self, query: str, docs_data: str, log_context: LogContext = None
-    ) -> Generator[str, None, None]:
-        plan_prompt_filled = planning_prompt_template.replace("{query}", query)
-        if "{summaries}" in plan_prompt_filled:
-            summaries = docs_data if docs_data else "No documents retrieved."
-            plan_prompt_filled = plan_prompt_filled.replace("{summaries}", summaries)
-        plan_prompt_filled = plan_prompt_filled.replace("{prompt}", self.prompt or "")
-        plan_prompt_filled = plan_prompt_filled.replace(
-            "{observations}", "\n".join(self.observations)
-        )
+            if satisfied:
+                logger.info("ReActAgent: Goal satisfied, stopping reasoning loop")
+                break
+        yield from self._synthesis_phase(query, log_context)
 
-        messages = [{"role": "user", "content": plan_prompt_filled}]
+    def _reset_state(self):
+        """Reset agent state for new query"""
+        self.plan = ""
+        self.observations = []
 
-        plan_stream_from_llm = self.llm.gen_stream(
-            model=self.gpt_model,
+    def _planning_phase(
+        self, query: str, log_context: LogContext
+    ) -> Generator[Dict, None, None]:
+        """Generate strategic plan for query"""
+        logger.info("ReActAgent: Creating plan...")
+
+        plan_prompt = self._build_planning_prompt(query)
+        messages = [{"role": "user", "content": plan_prompt}]
+
+        plan_stream = self.llm.gen_stream(
+            model=self.model_id,
             messages=messages,
-            tools=getattr(self, "tools", None),  # Use self.tools
+            tools=self.tools if self.tools else None,
         )
+
         if log_context:
-            data = build_stack_data(self.llm)
-            log_context.stacks.append({"component": "planning_llm", "data": data})
-
-        for chunk in plan_stream_from_llm:
-            content_piece = self._extract_content_from_llm_response(chunk)
-            if content_piece:
-                yield content_piece
-
-    def _create_final_answer(
-        self, query: str, observations: List[str], log_context: LogContext = None
-    ) -> Generator[str, None, None]:
-        observation_string = "\n".join(observations)
-        max_obs_len = 10000
-        if len(observation_string) > max_obs_len:
-            observation_string = (
-                observation_string[:max_obs_len] + "\n...[observations truncated]"
-            )
-            logger.warning(
-                "ReActAgent: Truncated observations for final answer prompt due to length."
+            log_context.stacks.append(
+                {"component": "planning_llm", "data": build_stack_data(self.llm)}
             )
+        plan_parts = []
+        for chunk in plan_stream:
+            content = self._extract_content(chunk)
+            if content:
+                plan_parts.append(content)
+                yield {"thought": content}
+        self.plan = "".join(plan_parts)
 
-        final_answer_prompt_filled = final_prompt_template.format(
-            query=query, observations=observation_string
+    def _execution_phase(
+        self, query: str, tools_dict: Dict, log_context: LogContext
+    ) -> Generator[bool, None, None]:
+        """Execute plan with tool calls and observations"""
+        execution_prompt = self._build_execution_prompt(query)
+        messages = self._build_messages(execution_prompt, query)
+
+        llm_response = self._llm_gen(messages, log_context)
+        initial_content = self._extract_content(llm_response)
+
+        if initial_content:
+            self.observations.append(f"Initial response: {initial_content}")
+        processed_response = self._llm_handler(
+            llm_response, tools_dict, messages, log_context
         )
 
-        messages = [{"role": "user", "content": final_answer_prompt_filled}]
-
-        # Final answer should synthesize, not call tools.
-        final_answer_stream_from_llm = self.llm.gen_stream(
-            model=self.gpt_model, messages=messages, tools=None
-        )
+        for tool_call in self.tool_calls:
+            observation = (
+                f"Executed: {tool_call.get('tool_name', 'Unknown')} "
+                f"with args {tool_call.get('arguments', {})}. "
+                f"Result: {str(tool_call.get('result', ''))[:200]}"
+            )
+            self.observations.append(observation)
+        final_content = self._extract_content(processed_response)
+        if final_content:
+            self.observations.append(f"Response after tools: {final_content}")
         if log_context:
-            data = build_stack_data(self.llm)
-            log_context.stacks.append({"component": "final_answer_llm", "data": data})
+            log_context.stacks.append(
+                {
+                    "component": "agent_tool_calls",
+                    "data": {"tool_calls": self.tool_calls.copy()},
+                }
+            )
+        yield {"sources": self.retrieved_docs}
+        yield {"tool_calls": self._get_truncated_tool_calls()}
 
-        for chunk in final_answer_stream_from_llm:
-            content_piece = self._extract_content_from_llm_response(chunk)
-            if content_piece:
-                yield content_piece
+        return "SATISFIED" in (final_content or "")
+
+    def _synthesis_phase(
+        self, query: str, log_context: LogContext
+    ) -> Generator[Dict, None, None]:
+        """Synthesize final answer from all observations"""
+        logger.info("ReActAgent: Generating final answer...")
+
+        final_prompt = self._build_final_answer_prompt(query)
+        messages = [{"role": "user", "content": final_prompt}]
+
+        final_stream = self.llm.gen_stream(
+            model=self.model_id, messages=messages, tools=None
+        )
+
+        if log_context:
+            log_context.stacks.append(
+                {"component": "final_answer_llm", "data": build_stack_data(self.llm)}
+            )
+        for chunk in final_stream:
+            content = self._extract_content(chunk)
+            if content:
+                yield {"answer": content}
+
+    def _build_planning_prompt(self, query: str) -> str:
+        """Build planning phase prompt"""
+        prompt = PLANNING_PROMPT_TEMPLATE.replace("{query}", query)
+        prompt = prompt.replace("{prompt}", self.prompt or "")
+        prompt = prompt.replace("{summaries}", "")
+        prompt = prompt.replace("{observations}", "\n".join(self.observations))
+        return prompt
+
+    def _build_execution_prompt(self, query: str) -> str:
+        """Build execution phase prompt with plan and observations"""
+        observations_str = "\n".join(self.observations)
+
+        if len(observations_str) > 20000:
+            observations_str = observations_str[:20000] + "\n...[truncated]"
+        return (
+            f"{self.prompt or ''}\n\n"
+            f"Follow this plan:\n{self.plan}\n\n"
+            f"Observations:\n{observations_str}\n\n"
+            f"If sufficient data exists to answer '{query}', respond with 'SATISFIED'. "
+            f"Otherwise, continue executing the plan."
+        )
+
+    def _build_final_answer_prompt(self, query: str) -> str:
+        """Build final synthesis prompt"""
+        observations_str = "\n".join(self.observations)
+
+        if len(observations_str) > 10000:
+            observations_str = observations_str[:10000] + "\n...[truncated]"
+            logger.warning("ReActAgent: Observations truncated for final answer")
+        return FINAL_PROMPT_TEMPLATE.format(query=query, observations=observations_str)
+
+    def _extract_content(self, response: Any) -> str:
+        """Extract text content from various LLM response formats"""
+        if not response:
+            return ""
+        collected = []
+
+        if isinstance(response, str):
+            return response
+        if hasattr(response, "message") and hasattr(response.message, "content"):
+            if response.message.content:
+                return response.message.content
+        if hasattr(response, "choices") and response.choices:
+            if hasattr(response.choices[0], "message"):
+                content = response.choices[0].message.content
+                if content:
+                    return content
+        if hasattr(response, "content") and isinstance(response.content, list):
+            if response.content and hasattr(response.content[0], "text"):
+                return response.content[0].text
+        try:
+            for chunk in response:
+                content_piece = ""
+
+                if hasattr(chunk, "choices") and chunk.choices:
+                    if hasattr(chunk.choices[0], "delta"):
+                        delta_content = chunk.choices[0].delta.content
+                        if delta_content:
+                            content_piece = delta_content
+                elif hasattr(chunk, "type") and chunk.type == "content_block_delta":
+                    if hasattr(chunk, "delta") and hasattr(chunk.delta, "text"):
+                        content_piece = chunk.delta.text
+                elif isinstance(chunk, str):
+                    content_piece = chunk
+                if content_piece:
+                    collected.append(content_piece)
+        except (TypeError, AttributeError):
+            logger.debug(
+                f"Response not iterable or unexpected format: {type(response)}"
+            )
+        except Exception as e:
+            logger.error(f"Error extracting content: {e}")
+        return "".join(collected)

application/api/answer/routes/answer.py

@@ -54,6 +54,14 @@ class AnswerResource(Resource, BaseAnswerResource):
                 default=True,
                 description="Whether to save the conversation",
             ),
+            "model_id": fields.String(
+                required=False,
+                description="Model ID to use for this request",
+            ),
+            "passthrough": fields.Raw(
+                required=False,
+                description="Dynamic parameters to inject into prompt template",
+            ),
         },
     )
 
@@ -69,8 +77,17 @@ class AnswerResource(Resource, BaseAnswerResource):
             processor.initialize()
             if not processor.decoded_token:
                 return make_response({"error": "Unauthorized"}, 401)
-            agent = processor.create_agent()
-            retriever = processor.create_retriever()
+
+            docs_together, docs_list = processor.pre_fetch_docs(
+                data.get("question", "")
+            )
+            tools_data = processor.pre_fetch_tools()
+
+            agent = processor.create_agent(
+                docs_together=docs_together,
+                docs=docs_list,
+                tools_data=tools_data,
+            )
 
             if error := self.check_usage(processor.agent_config):
                 return error
@@ -78,13 +95,13 @@ class AnswerResource(Resource, BaseAnswerResource):
             stream = self.complete_stream(
                 question=data["question"],
                 agent=agent,
-                retriever=retriever,
                 conversation_id=processor.conversation_id,
                 user_api_key=processor.agent_config.get("user_api_key"),
                 decoded_token=processor.decoded_token,
                 isNoneDoc=data.get("isNoneDoc"),
                 index=None,
                 should_save_conversation=data.get("save_conversation", True),
+                model_id=processor.model_id,
             )
             stream_result = self.process_response_stream(stream)
 

application/api/answer/routes/base.py

@@ -3,15 +3,20 @@ import json
 import logging
 from typing import Any, Dict, Generator, List, Optional
 
-from flask import Response, make_response, jsonify
+from flask import jsonify, make_response, Response
 from flask_restx import Namespace
 
 from application.api.answer.services.conversation_service import ConversationService
+from application.core.model_utils import (
+    get_api_key_for_provider,
+    get_default_model_id,
+    get_provider_from_model_id,
+)
 
 from application.core.mongo_db import MongoDB
 from application.core.settings import settings
 from application.llm.llm_creator import LLMCreator
-from application.utils import check_required_fields, get_gpt_model
+from application.utils import check_required_fields
 
 logger = logging.getLogger(__name__)
 
@@ -27,7 +32,7 @@ class BaseAnswerResource:
         db = mongo[settings.MONGO_DB_NAME]
         self.db = db
         self.user_logs_collection = db["user_logs"]
-        self.gpt_model = get_gpt_model()
+        self.default_model_id = get_default_model_id()
         self.conversation_service = ConversationService()
 
     def validate_request(
@@ -41,9 +46,7 @@ class BaseAnswerResource:
             return missing_fields
         return None
 
-    def check_usage(
-            self, agent_config: Dict
-    ) -> Optional[Response]:
+    def check_usage(self, agent_config: Dict) -> Optional[Response]:
         """Check if there is a usage limit and if it is exceeded
 
         Args:
@@ -51,30 +54,38 @@ class BaseAnswerResource:
 
         Returns:
             None or Response if either of limits exceeded.
-        
+
         """
         api_key = agent_config.get("user_api_key")
         if not api_key:
             return None
-        
         agents_collection = self.db["agents"]
         agent = agents_collection.find_one({"key": api_key})
 
         if not agent:
             return make_response(
-                jsonify(
-                    {
-                        "success": False,
-                        "message": "Invalid API key."
-                    }
-                ),
-                401
+                jsonify({"success": False, "message": "Invalid API key."}), 401
             )
+        limited_token_mode_raw = agent.get("limited_token_mode", False)
+        limited_request_mode_raw = agent.get("limited_request_mode", False)
 
-        limited_token_mode = agent.get("limited_token_mode", False)
-        limited_request_mode = agent.get("limited_request_mode", False)
-        token_limit = int(agent.get("token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]))
-        request_limit = int(agent.get("request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]))
+        limited_token_mode = (
+            limited_token_mode_raw
+            if isinstance(limited_token_mode_raw, bool)
+            else limited_token_mode_raw == "True"
+        )
+        limited_request_mode = (
+            limited_request_mode_raw
+            if isinstance(limited_request_mode_raw, bool)
+            else limited_request_mode_raw == "True"
+        )
+
+        token_limit = int(
+            agent.get("token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"])
+        )
+        request_limit = int(
+            agent.get("request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"])
+        )
 
         token_usage_collection = self.db["token_usage"]
 
@@ -83,51 +94,56 @@ class BaseAnswerResource:
 
         match_query = {
             "timestamp": {"$gte": start_date, "$lte": end_date},
-            "api_key": api_key
+            "api_key": api_key,
         }
-        
+
         if limited_token_mode:
             token_pipeline = [
                 {"$match": match_query},
                 {
                     "$group": {
                         "_id": None,
-                        "total_tokens": {"$sum": {"$add": ["$prompt_tokens", "$generated_tokens"]}}
+                        "total_tokens": {
+                            "$sum": {"$add": ["$prompt_tokens", "$generated_tokens"]}
+                        },
                     }
-                }
+                },
             ]
             token_result = list(token_usage_collection.aggregate(token_pipeline))
             daily_token_usage = token_result[0]["total_tokens"] if token_result else 0
         else:
             daily_token_usage = 0
-
         if limited_request_mode:
             daily_request_usage = token_usage_collection.count_documents(match_query)
         else:
             daily_request_usage = 0
-
         if not limited_token_mode and not limited_request_mode:
             return None
-        elif limited_token_mode and token_limit > daily_token_usage:
-            return None
-        elif limited_request_mode and request_limit > daily_request_usage:
-            return None
-
-        return make_response(
-            jsonify(
-                {
-                    "success": False,
-                    "message": "Exceeding usage limit, please try again later."
-                }
-            ),
-            429, # too many requests
+        token_exceeded = (
+            limited_token_mode and token_limit > 0 and daily_token_usage >= token_limit
         )
+        request_exceeded = (
+            limited_request_mode
+            and request_limit > 0
+            and daily_request_usage >= request_limit
+        )
+
+        if token_exceeded or request_exceeded:
+            return make_response(
+                jsonify(
+                    {
+                        "success": False,
+                        "message": "Exceeding usage limit, please try again later.",
+                    }
+                ),
+                429,
+            )
+        return None
 
     def complete_stream(
         self,
         question: str,
         agent: Any,
-        retriever: Any,
         conversation_id: Optional[str],
         user_api_key: Optional[str],
         decoded_token: Dict[str, Any],
@@ -138,6 +154,7 @@ class BaseAnswerResource:
         agent_id: Optional[str] = None,
         is_shared_usage: bool = False,
         shared_token: Optional[str] = None,
+        model_id: Optional[str] = None,
     ) -> Generator[str, None, None]:
         """
         Generator function that streams the complete conversation response.
@@ -156,6 +173,8 @@ class BaseAnswerResource:
             agent_id: ID of agent used
             is_shared_usage: Flag for shared agent usage
             shared_token: Token for shared agent
+            model_id: Model ID used for the request
+            retrieved_docs: Pre-fetched documents for sources (optional)
 
         Yields:
             Server-sent event strings
@@ -166,7 +185,7 @@ class BaseAnswerResource:
             schema_info = None
             structured_chunks = []
 
-            for line in agent.gen(query=question, retriever=retriever):
+            for line in agent.gen(query=question):
                 if "answer" in line:
                     response_full += str(line["answer"])
                     if line.get("structured"):
@@ -202,7 +221,6 @@ class BaseAnswerResource:
                 elif "type" in line:
                     data = json.dumps(line)
                     yield f"data: {data}\n\n"
-
             if is_structured and structured_chunks:
                 structured_data = {
                     "type": "structured_answer",
@@ -212,15 +230,22 @@ class BaseAnswerResource:
                 }
                 data = json.dumps(structured_data)
                 yield f"data: {data}\n\n"
-
             if isNoneDoc:
                 for doc in source_log_docs:
                     doc["source"] = "None"
+            provider = (
+                get_provider_from_model_id(model_id)
+                if model_id
+                else settings.LLM_PROVIDER
+            )
+            system_api_key = get_api_key_for_provider(provider or settings.LLM_PROVIDER)
+
             llm = LLMCreator.create_llm(
-                settings.LLM_PROVIDER,
-                api_key=settings.API_KEY,
+                provider or settings.LLM_PROVIDER,
+                api_key=system_api_key,
                 user_api_key=user_api_key,
                 decoded_token=decoded_token,
+                model_id=model_id,
             )
 
             if should_save_conversation:
@@ -232,7 +257,7 @@ class BaseAnswerResource:
                     source_log_docs,
                     tool_calls,
                     llm,
-                    self.gpt_model,
+                    model_id or self.default_model_id,
                     decoded_token,
                     index=index,
                     api_key=user_api_key,
@@ -241,13 +266,32 @@ class BaseAnswerResource:
                     shared_token=shared_token,
                     attachment_ids=attachment_ids,
                 )
+                # Persist compression metadata/summary if it exists and wasn't saved mid-execution
+                compression_meta = getattr(agent, "compression_metadata", None)
+                compression_saved = getattr(agent, "compression_saved", False)
+                if conversation_id and compression_meta and not compression_saved:
+                    try:
+                        self.conversation_service.update_compression_metadata(
+                            conversation_id, compression_meta
+                        )
+                        self.conversation_service.append_compression_message(
+                            conversation_id, compression_meta
+                        )
+                        agent.compression_saved = True
+                        logger.info(
+                            f"Persisted compression metadata for conversation {conversation_id}"
+                        )
+                    except Exception as e:
+                        logger.error(
+                            f"Failed to persist compression metadata: {str(e)}",
+                            exc_info=True,
+                        )
             else:
                 conversation_id = None
             id_data = {"type": "id", "id": str(conversation_id)}
             data = json.dumps(id_data)
             yield f"data: {data}\n\n"
 
-            retriever_params = retriever.get_params()
             log_data = {
                 "action": "stream_answer",
                 "level": "info",
@@ -256,7 +300,6 @@ class BaseAnswerResource:
                 "question": question,
                 "response": response_full,
                 "sources": source_log_docs,
-                "retriever_params": retriever_params,
                 "attachments": attachment_ids,
                 "timestamp": datetime.datetime.now(datetime.timezone.utc),
             }
@@ -264,24 +307,19 @@ class BaseAnswerResource:
                 log_data["structured_output"] = True
                 if schema_info:
                     log_data["schema"] = schema_info
-  
-            # clean up text fields to be no longer than 10000 characters
+            # Clean up text fields to be no longer than 10000 characters
+
             for key, value in log_data.items():
                 if isinstance(value, str) and len(value) > 10000:
                     log_data[key] = value[:10000]
-            
             self.user_logs_collection.insert_one(log_data)
 
-            # End of stream
-
             data = json.dumps({"type": "end"})
             yield f"data: {data}\n\n"
         except GeneratorExit:
-            # Client aborted the connection
-            logger.info(
-                f"Stream aborted by client for question: {question[:50]}... "
-            )
-            # Save partial response to database before exiting
+            logger.info(f"Stream aborted by client for question: {question[:50]}... ")
+            # Save partial response
+
             if should_save_conversation and response_full:
                 try:
                     if isNoneDoc:
@@ -301,7 +339,7 @@ class BaseAnswerResource:
                         source_log_docs,
                         tool_calls,
                         llm,
-                        self.gpt_model,
+                        model_id or self.default_model_id,
                         decoded_token,
                         index=index,
                         api_key=user_api_key,
@@ -310,8 +348,29 @@ class BaseAnswerResource:
                         shared_token=shared_token,
                         attachment_ids=attachment_ids,
                     )
+                    compression_meta = getattr(agent, "compression_metadata", None)
+                    compression_saved = getattr(agent, "compression_saved", False)
+                    if conversation_id and compression_meta and not compression_saved:
+                        try:
+                            self.conversation_service.update_compression_metadata(
+                                conversation_id, compression_meta
+                            )
+                            self.conversation_service.append_compression_message(
+                                conversation_id, compression_meta
+                            )
+                            agent.compression_saved = True
+                            logger.info(
+                                f"Persisted compression metadata for conversation {conversation_id} (partial stream)"
+                            )
+                        except Exception as e:
+                            logger.error(
+                                f"Failed to persist compression metadata (partial stream): {str(e)}",
+                                exc_info=True,
+                            )
                 except Exception as e:
-                    logger.error(f"Error saving partial response: {str(e)}", exc_info=True)
+                    logger.error(
+                        f"Error saving partial response: {str(e)}", exc_info=True
+                    )
             raise
         except Exception as e:
             logger.error(f"Error in stream: {str(e)}", exc_info=True)
@@ -356,7 +415,7 @@ class BaseAnswerResource:
                     thought = event["thought"]
                 elif event["type"] == "error":
                     logger.error(f"Error from stream: {event['error']}")
-                    return None, None, None, None, event["error"]
+                    return None, None, None, None, event["error"], None
                 elif event["type"] == "end":
                     stream_ended = True
             except (json.JSONDecodeError, KeyError) as e:
@@ -364,8 +423,7 @@ class BaseAnswerResource:
                 continue
         if not stream_ended:
             logger.error("Stream ended unexpectedly without an 'end' event.")
-            return None, None, None, None, "Stream ended unexpectedly"
-
+            return None, None, None, None, "Stream ended unexpectedly", None
         result = (
             conversation_id,
             response_full,
@@ -377,7 +435,6 @@ class BaseAnswerResource:
 
         if is_structured:
             result = result + ({"structured": True, "schema": schema_info},)
-
         return result
 
     def error_stream_generate(self, err_response):

application/api/answer/routes/stream.py

@@ -57,9 +57,17 @@ class StreamResource(Resource, BaseAnswerResource):
                 default=True,
                 description="Whether to save the conversation",
             ),
+            "model_id": fields.String(
+                required=False,
+                description="Model ID to use for this request",
+            ),
             "attachments": fields.List(
                 fields.String, required=False, description="List of attachment IDs"
             ),
+            "passthrough": fields.Raw(
+                required=False,
+                description="Dynamic parameters to inject into prompt template",
+            ),
         },
     )
 
@@ -73,17 +81,20 @@ class StreamResource(Resource, BaseAnswerResource):
         processor = StreamProcessor(data, decoded_token)
         try:
             processor.initialize()
-            agent = processor.create_agent()
-            retriever = processor.create_retriever()
+
+            docs_together, docs_list = processor.pre_fetch_docs(data["question"])
+            tools_data = processor.pre_fetch_tools()
+
+            agent = processor.create_agent(
+                docs_together=docs_together, docs=docs_list, tools_data=tools_data
+            )
 
             if error := self.check_usage(processor.agent_config):
                 return error
-
             return Response(
                 self.complete_stream(
                     question=data["question"],
                     agent=agent,
-                    retriever=retriever,
                     conversation_id=processor.conversation_id,
                     user_api_key=processor.agent_config.get("user_api_key"),
                     decoded_token=processor.decoded_token,
@@ -94,6 +105,7 @@ class StreamResource(Resource, BaseAnswerResource):
                     agent_id=data.get("agent_id"),
                     is_shared_usage=processor.is_shared_usage,
                     shared_token=processor.shared_token,
+                    model_id=processor.model_id,
                 ),
                 mimetype="text/event-stream",
             )

application/api/answer/services/compression/__init__.py

@@ -0,0 +1,20 @@
+"""
+Compression module for managing conversation context compression.
+
+"""
+
+from application.api.answer.services.compression.orchestrator import (
+    CompressionOrchestrator,
+)
+from application.api.answer.services.compression.service import CompressionService
+from application.api.answer.services.compression.types import (
+    CompressionResult,
+    CompressionMetadata,
+)
+
+__all__ = [
+    "CompressionOrchestrator",
+    "CompressionService",
+    "CompressionResult",
+    "CompressionMetadata",
+]

application/api/answer/services/compression/message_builder.py

@@ -0,0 +1,234 @@
+"""Message reconstruction utilities for compression."""
+
+import logging
+import uuid
+from typing import Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+class MessageBuilder:
+    """Builds message arrays from compressed context."""
+
+    @staticmethod
+    def build_from_compressed_context(
+        system_prompt: str,
+        compressed_summary: Optional[str],
+        recent_queries: List[Dict],
+        include_tool_calls: bool = False,
+        context_type: str = "pre_request",
+    ) -> List[Dict]:
+        """
+        Build messages from compressed context.
+
+        Args:
+            system_prompt: Original system prompt
+            compressed_summary: Compressed summary (if any)
+            recent_queries: Recent uncompressed queries
+            include_tool_calls: Whether to include tool calls from history
+            context_type: Type of context ('pre_request' or 'mid_execution')
+
+        Returns:
+            List of message dicts ready for LLM
+        """
+        # Append compression summary to system prompt if present
+        if compressed_summary:
+            system_prompt = MessageBuilder._append_compression_context(
+                system_prompt, compressed_summary, context_type
+            )
+
+        messages = [{"role": "system", "content": system_prompt}]
+
+        # Add recent history
+        for query in recent_queries:
+            if "prompt" in query and "response" in query:
+                messages.append({"role": "user", "content": query["prompt"]})
+                messages.append({"role": "assistant", "content": query["response"]})
+
+            # Add tool calls from history if present
+            if include_tool_calls and "tool_calls" in query:
+                for tool_call in query["tool_calls"]:
+                    call_id = tool_call.get("call_id") or str(uuid.uuid4())
+
+                    function_call_dict = {
+                        "function_call": {
+                            "name": tool_call.get("action_name"),
+                            "args": tool_call.get("arguments"),
+                            "call_id": call_id,
+                        }
+                    }
+                    function_response_dict = {
+                        "function_response": {
+                            "name": tool_call.get("action_name"),
+                            "response": {"result": tool_call.get("result")},
+                            "call_id": call_id,
+                        }
+                    }
+
+                    messages.append(
+                        {"role": "assistant", "content": [function_call_dict]}
+                    )
+                    messages.append(
+                        {"role": "tool", "content": [function_response_dict]}
+                    )
+
+        # If no recent queries (everything was compressed), add a continuation user message
+        if len(recent_queries) == 0 and compressed_summary:
+            messages.append({
+                "role": "user",
+                "content": "Please continue with the remaining tasks based on the context above."
+            })
+            logger.info("Added continuation user message to maintain proper turn-taking after full compression")
+
+        return messages
+
+    @staticmethod
+    def _append_compression_context(
+        system_prompt: str, compressed_summary: str, context_type: str = "pre_request"
+    ) -> str:
+        """
+        Append compression context to system prompt.
+
+        Args:
+            system_prompt: Original system prompt
+            compressed_summary: Summary to append
+            context_type: Type of compression context
+
+        Returns:
+            Updated system prompt
+        """
+        # Remove existing compression context if present
+        if "This session is being continued" in system_prompt or "Context window limit reached" in system_prompt:
+            parts = system_prompt.split("\n\n---\n\n")
+            system_prompt = parts[0]
+
+        # Build appropriate context message based on type
+        if context_type == "mid_execution":
+            context_message = (
+                "\n\n---\n\n"
+                "Context window limit reached during execution. "
+                "Previous conversation has been compressed to fit within limits. "
+                "The conversation is summarized below:\n\n"
+                f"{compressed_summary}"
+            )
+        else:  # pre_request
+            context_message = (
+                "\n\n---\n\n"
+                "This session is being continued from a previous conversation that "
+                "has been compressed to fit within context limits. "
+                "The conversation is summarized below:\n\n"
+                f"{compressed_summary}"
+            )
+
+        return system_prompt + context_message
+
+    @staticmethod
+    def rebuild_messages_after_compression(
+        messages: List[Dict],
+        compressed_summary: Optional[str],
+        recent_queries: List[Dict],
+        include_current_execution: bool = False,
+        include_tool_calls: bool = False,
+    ) -> Optional[List[Dict]]:
+        """
+        Rebuild the message list after compression so tool execution can continue.
+
+        Args:
+            messages: Original message list
+            compressed_summary: Compressed summary
+            recent_queries: Recent uncompressed queries
+            include_current_execution: Whether to preserve current execution messages
+            include_tool_calls: Whether to include tool calls from history
+
+        Returns:
+            Rebuilt message list or None if failed
+        """
+        # Find the system message
+        system_message = next(
+            (msg for msg in messages if msg.get("role") == "system"), None
+        )
+        if not system_message:
+            logger.warning("No system message found in messages list")
+            return None
+
+        # Update system message with compressed summary
+        if compressed_summary:
+            content = system_message.get("content", "")
+            system_message["content"] = MessageBuilder._append_compression_context(
+                content, compressed_summary, "mid_execution"
+            )
+            logger.info(
+                "Appended compression summary to system prompt (truncated): %s",
+                (
+                    compressed_summary[:500] + "..."
+                    if len(compressed_summary) > 500
+                    else compressed_summary
+                ),
+            )
+
+        rebuilt_messages = [system_message]
+
+        # Add recent history from compressed context
+        for query in recent_queries:
+            if "prompt" in query and "response" in query:
+                rebuilt_messages.append({"role": "user", "content": query["prompt"]})
+                rebuilt_messages.append(
+                    {"role": "assistant", "content": query["response"]}
+                )
+
+            # Add tool calls from history if present
+            if include_tool_calls and "tool_calls" in query:
+                for tool_call in query["tool_calls"]:
+                    call_id = tool_call.get("call_id") or str(uuid.uuid4())
+
+                    function_call_dict = {
+                        "function_call": {
+                            "name": tool_call.get("action_name"),
+                            "args": tool_call.get("arguments"),
+                            "call_id": call_id,
+                        }
+                    }
+                    function_response_dict = {
+                        "function_response": {
+                            "name": tool_call.get("action_name"),
+                            "response": {"result": tool_call.get("result")},
+                            "call_id": call_id,
+                        }
+                    }
+
+                    rebuilt_messages.append(
+                        {"role": "assistant", "content": [function_call_dict]}
+                    )
+                    rebuilt_messages.append(
+                        {"role": "tool", "content": [function_response_dict]}
+                    )
+
+        # If no recent queries (everything was compressed), add a continuation user message
+        if len(recent_queries) == 0 and compressed_summary:
+            rebuilt_messages.append({
+                "role": "user",
+                "content": "Please continue with the remaining tasks based on the context above."
+            })
+            logger.info("Added continuation user message to maintain proper turn-taking after full compression")
+
+        if include_current_execution:
+            # Preserve any messages that were added during the current execution cycle
+            recent_msg_count = 1  # system message
+            for query in recent_queries:
+                if "prompt" in query and "response" in query:
+                    recent_msg_count += 2
+                if "tool_calls" in query:
+                    recent_msg_count += len(query["tool_calls"]) * 2
+
+            if len(messages) > recent_msg_count:
+                current_execution_messages = messages[recent_msg_count:]
+                rebuilt_messages.extend(current_execution_messages)
+                logger.info(
+                    f"Preserved {len(current_execution_messages)} messages from current execution cycle"
+                )
+
+        logger.info(
+            f"Messages rebuilt: {len(messages)} → {len(rebuilt_messages)} messages. "
+            f"Ready to continue tool execution."
+        )
+        return rebuilt_messages

application/api/answer/services/compression/orchestrator.py

@@ -0,0 +1,232 @@
+"""High-level compression orchestration."""
+
+import logging
+from typing import Any, Dict, Optional
+
+from application.api.answer.services.compression.service import CompressionService
+from application.api.answer.services.compression.threshold_checker import (
+    CompressionThresholdChecker,
+)
+from application.api.answer.services.compression.types import CompressionResult
+from application.api.answer.services.conversation_service import ConversationService
+from application.core.model_utils import (
+    get_api_key_for_provider,
+    get_provider_from_model_id,
+)
+from application.core.settings import settings
+from application.llm.llm_creator import LLMCreator
+
+logger = logging.getLogger(__name__)
+
+
+class CompressionOrchestrator:
+    """
+    Facade for compression operations.
+
+    Coordinates between all compression components and provides
+    a simple interface for callers.
+    """
+
+    def __init__(
+        self,
+        conversation_service: ConversationService,
+        threshold_checker: Optional[CompressionThresholdChecker] = None,
+    ):
+        """
+        Initialize orchestrator.
+
+        Args:
+            conversation_service: Service for DB operations
+            threshold_checker: Custom threshold checker (optional)
+        """
+        self.conversation_service = conversation_service
+        self.threshold_checker = threshold_checker or CompressionThresholdChecker()
+
+    def compress_if_needed(
+        self,
+        conversation_id: str,
+        user_id: str,
+        model_id: str,
+        decoded_token: Dict[str, Any],
+        current_query_tokens: int = 500,
+    ) -> CompressionResult:
+        """
+        Check if compression is needed and perform it if so.
+
+        This is the main entry point for compression operations.
+
+        Args:
+            conversation_id: Conversation ID
+            user_id: User ID
+            model_id: Model being used for conversation
+            decoded_token: User's decoded JWT token
+            current_query_tokens: Estimated tokens for current query
+
+        Returns:
+            CompressionResult with summary and recent queries
+        """
+        try:
+            # Load conversation
+            conversation = self.conversation_service.get_conversation(
+                conversation_id, user_id
+            )
+
+            if not conversation:
+                logger.warning(
+                    f"Conversation {conversation_id} not found for user {user_id}"
+                )
+                return CompressionResult.failure("Conversation not found")
+
+            # Check if compression is needed
+            if not self.threshold_checker.should_compress(
+                conversation, model_id, current_query_tokens
+            ):
+                # No compression needed, return full history
+                queries = conversation.get("queries", [])
+                return CompressionResult.success_no_compression(queries)
+
+            # Perform compression
+            return self._perform_compression(
+                conversation_id, conversation, model_id, decoded_token
+            )
+
+        except Exception as e:
+            logger.error(
+                f"Error in compress_if_needed: {str(e)}", exc_info=True
+            )
+            return CompressionResult.failure(str(e))
+
+    def _perform_compression(
+        self,
+        conversation_id: str,
+        conversation: Dict[str, Any],
+        model_id: str,
+        decoded_token: Dict[str, Any],
+    ) -> CompressionResult:
+        """
+        Perform the actual compression operation.
+
+        Args:
+            conversation_id: Conversation ID
+            conversation: Conversation document
+            model_id: Model ID for conversation
+            decoded_token: User token
+
+        Returns:
+            CompressionResult
+        """
+        try:
+            # Determine which model to use for compression
+            compression_model = (
+                settings.COMPRESSION_MODEL_OVERRIDE
+                if settings.COMPRESSION_MODEL_OVERRIDE
+                else model_id
+            )
+
+            # Get provider and API key for compression model
+            provider = get_provider_from_model_id(compression_model)
+            api_key = get_api_key_for_provider(provider)
+
+            # Create compression LLM
+            compression_llm = LLMCreator.create_llm(
+                provider,
+                api_key=api_key,
+                user_api_key=None,
+                decoded_token=decoded_token,
+                model_id=compression_model,
+            )
+
+            # Create compression service with DB update capability
+            compression_service = CompressionService(
+                llm=compression_llm,
+                model_id=compression_model,
+                conversation_service=self.conversation_service,
+            )
+
+            # Compress all queries up to the latest
+            queries_count = len(conversation.get("queries", []))
+            compress_up_to = queries_count - 1
+
+            if compress_up_to < 0:
+                logger.warning("No queries to compress")
+                return CompressionResult.success_no_compression([])
+
+            logger.info(
+                f"Initiating compression for conversation {conversation_id}: "
+                f"compressing all {queries_count} queries (0-{compress_up_to})"
+            )
+
+            # Perform compression and save to DB
+            metadata = compression_service.compress_and_save(
+                conversation_id, conversation, compress_up_to
+            )
+
+            logger.info(
+                f"Compression successful - ratio: {metadata.compression_ratio:.1f}x, "
+                f"saved {metadata.original_token_count - metadata.compressed_token_count} tokens"
+            )
+
+            # Reload conversation with updated metadata
+            conversation = self.conversation_service.get_conversation(
+                conversation_id, user_id=decoded_token.get("sub")
+            )
+
+            # Get compressed context
+            compressed_summary, recent_queries = (
+                compression_service.get_compressed_context(conversation)
+            )
+
+            return CompressionResult.success_with_compression(
+                compressed_summary, recent_queries, metadata
+            )
+
+        except Exception as e:
+            logger.error(f"Error performing compression: {str(e)}", exc_info=True)
+            return CompressionResult.failure(str(e))
+
+    def compress_mid_execution(
+        self,
+        conversation_id: str,
+        user_id: str,
+        model_id: str,
+        decoded_token: Dict[str, Any],
+        current_conversation: Optional[Dict[str, Any]] = None,
+    ) -> CompressionResult:
+        """
+        Perform compression during tool execution.
+
+        Args:
+            conversation_id: Conversation ID
+            user_id: User ID
+            model_id: Model ID
+            decoded_token: User token
+            current_conversation: Pre-loaded conversation (optional)
+
+        Returns:
+            CompressionResult
+        """
+        try:
+            # Load conversation if not provided
+            if current_conversation:
+                conversation = current_conversation
+            else:
+                conversation = self.conversation_service.get_conversation(
+                    conversation_id, user_id
+                )
+
+            if not conversation:
+                logger.warning(
+                    f"Could not load conversation {conversation_id} for mid-execution compression"
+                )
+                return CompressionResult.failure("Conversation not found")
+
+            # Perform compression
+            return self._perform_compression(
+                conversation_id, conversation, model_id, decoded_token
+            )
+
+        except Exception as e:
+            logger.error(
+                f"Error in mid-execution compression: {str(e)}", exc_info=True
+            )
+            return CompressionResult.failure(str(e))

application/api/answer/services/compression/prompt_builder.py

@@ -0,0 +1,149 @@
+"""Compression prompt building logic."""
+
+import logging
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+class CompressionPromptBuilder:
+    """Builds prompts for LLM compression calls."""
+
+    def __init__(self, version: str = "v1.0"):
+        """
+        Initialize prompt builder.
+
+        Args:
+            version: Prompt template version to use
+        """
+        self.version = version
+        self.system_prompt = self._load_prompt(version)
+
+    def _load_prompt(self, version: str) -> str:
+        """
+        Load prompt template from file.
+
+        Args:
+            version: Version string (e.g., 'v1.0')
+
+        Returns:
+            Prompt template content
+
+        Raises:
+            FileNotFoundError: If prompt template file doesn't exist
+        """
+        current_dir = Path(__file__).resolve().parents[4]
+        prompt_path = current_dir / "prompts" / "compression" / f"{version}.txt"
+
+        try:
+            with open(prompt_path, "r") as f:
+                return f.read()
+        except FileNotFoundError:
+            logger.error(f"Compression prompt template not found: {prompt_path}")
+            raise FileNotFoundError(
+                f"Compression prompt template '{version}' not found at {prompt_path}. "
+                f"Please ensure the template file exists."
+            )
+
+    def build_prompt(
+        self,
+        queries: List[Dict[str, Any]],
+        existing_compressions: Optional[List[Dict[str, Any]]] = None,
+    ) -> List[Dict[str, str]]:
+        """
+        Build messages for compression LLM call.
+
+        Args:
+            queries: List of query objects to compress
+            existing_compressions: List of previous compression points
+
+        Returns:
+            List of message dicts for LLM
+        """
+        # Build conversation text
+        conversation_text = self._format_conversation(queries)
+
+        # Add existing compression context if present
+        existing_compression_context = ""
+        if existing_compressions and len(existing_compressions) > 0:
+            existing_compression_context = (
+                "\n\nIMPORTANT: This conversation has been compressed before. "
+                "Previous compression summaries:\n\n"
+            )
+            for i, comp in enumerate(existing_compressions):
+                existing_compression_context += (
+                    f"--- Compression {i + 1} (up to message {comp.get('query_index', 'unknown')}) ---\n"
+                    f"{comp.get('compressed_summary', '')}\n\n"
+                )
+            existing_compression_context += (
+                "Your task is to create a NEW summary that incorporates the context from "
+                "previous compressions AND the new messages below. The final summary should "
+                "be comprehensive and include all important information from both previous "
+                "compressions and new messages.\n\n"
+            )
+
+        user_prompt = (
+            f"{existing_compression_context}"
+            f"Here is the conversation to summarize:\n\n"
+            f"{conversation_text}"
+        )
+
+        messages = [
+            {"role": "system", "content": self.system_prompt},
+            {"role": "user", "content": user_prompt},
+        ]
+
+        return messages
+
+    def _format_conversation(self, queries: List[Dict[str, Any]]) -> str:
+        """
+        Format conversation queries into readable text for compression.
+
+        Args:
+            queries: List of query objects
+
+        Returns:
+            Formatted conversation text
+        """
+        conversation_lines = []
+
+        for i, query in enumerate(queries):
+            conversation_lines.append(f"--- Message {i + 1} ---")
+            conversation_lines.append(f"User: {query.get('prompt', '')}")
+
+            # Add tool calls if present
+            tool_calls = query.get("tool_calls", [])
+            if tool_calls:
+                conversation_lines.append("\nTool Calls:")
+                for tc in tool_calls:
+                    tool_name = tc.get("tool_name", "unknown")
+                    action_name = tc.get("action_name", "unknown")
+                    arguments = tc.get("arguments", {})
+                    result = tc.get("result", "")
+                    if result is None:
+                        result = ""
+                    status = tc.get("status", "unknown")
+
+                    # Include full tool result for complete compression context
+                    conversation_lines.append(
+                        f"  - {tool_name}.{action_name}({arguments}) "
+                        f"[{status}] → {result}"
+                    )
+
+            # Add agent thought if present
+            thought = query.get("thought", "")
+            if thought:
+                conversation_lines.append(f"\nAgent Thought: {thought}")
+
+            # Add assistant response
+            conversation_lines.append(f"\nAssistant: {query.get('response', '')}")
+
+            # Add sources if present
+            sources = query.get("sources", [])
+            if sources:
+                conversation_lines.append(f"\nSources Used: {len(sources)} documents")
+
+            conversation_lines.append("")  # Empty line between messages
+
+        return "\n".join(conversation_lines)

application/api/answer/services/compression/service.py

@@ -0,0 +1,306 @@
+"""Core compression service with simplified responsibilities."""
+
+import logging
+import re
+from datetime import datetime, timezone
+from typing import Any, Dict, List, Optional
+
+from application.api.answer.services.compression.prompt_builder import (
+    CompressionPromptBuilder,
+)
+from application.api.answer.services.compression.token_counter import TokenCounter
+from application.api.answer.services.compression.types import (
+    CompressionMetadata,
+)
+from application.core.settings import settings
+
+logger = logging.getLogger(__name__)
+
+
+class CompressionService:
+    """
+    Service for compressing conversation history.
+
+    Handles DB updates.
+    """
+
+    def __init__(
+        self,
+        llm,
+        model_id: str,
+        conversation_service=None,
+        prompt_builder: Optional[CompressionPromptBuilder] = None,
+    ):
+        """
+        Initialize compression service.
+
+        Args:
+            llm: LLM instance to use for compression
+            model_id: Model ID for compression
+            conversation_service: Service for DB operations (optional, for DB updates)
+            prompt_builder: Custom prompt builder (optional)
+        """
+        self.llm = llm
+        self.model_id = model_id
+        self.conversation_service = conversation_service
+        self.prompt_builder = prompt_builder or CompressionPromptBuilder(
+            version=settings.COMPRESSION_PROMPT_VERSION
+        )
+
+    def compress_conversation(
+        self,
+        conversation: Dict[str, Any],
+        compress_up_to_index: int,
+    ) -> CompressionMetadata:
+        """
+        Compress conversation history up to specified index.
+
+        Args:
+            conversation: Full conversation document
+            compress_up_to_index: Last query index to include in compression
+
+        Returns:
+            CompressionMetadata with compression details
+
+        Raises:
+            ValueError: If compress_up_to_index is invalid
+        """
+        try:
+            queries = conversation.get("queries", [])
+
+            if compress_up_to_index < 0 or compress_up_to_index >= len(queries):
+                raise ValueError(
+                    f"Invalid compress_up_to_index: {compress_up_to_index} "
+                    f"(conversation has {len(queries)} queries)"
+                )
+
+            # Get queries to compress
+            queries_to_compress = queries[: compress_up_to_index + 1]
+
+            # Check if there are existing compressions
+            existing_compressions = conversation.get("compression_metadata", {}).get(
+                "compression_points", []
+            )
+
+            if existing_compressions:
+                logger.info(
+                    f"Found {len(existing_compressions)} previous compression(s) - "
+                    f"will incorporate into new summary"
+                )
+
+            # Calculate original token count
+            original_tokens = TokenCounter.count_query_tokens(queries_to_compress)
+
+            # Log tool call stats
+            self._log_tool_call_stats(queries_to_compress)
+
+            # Build compression prompt
+            messages = self.prompt_builder.build_prompt(
+                queries_to_compress, existing_compressions
+            )
+
+            # Call LLM to generate compression
+            logger.info(
+                f"Starting compression: {len(queries_to_compress)} queries "
+                f"(messages 0-{compress_up_to_index}, {original_tokens} tokens) "
+                f"using model {self.model_id}"
+            )
+
+            response = self.llm.gen(
+                model=self.model_id, messages=messages, max_tokens=4000
+            )
+
+            # Extract summary from response
+            compressed_summary = self._extract_summary(response)
+
+            # Calculate compressed token count
+            compressed_tokens = TokenCounter.count_message_tokens(
+                [{"content": compressed_summary}]
+            )
+
+            # Calculate compression ratio
+            compression_ratio = (
+                original_tokens / compressed_tokens if compressed_tokens > 0 else 0
+            )
+
+            logger.info(
+                f"Compression complete: {original_tokens} → {compressed_tokens} tokens "
+                f"({compression_ratio:.1f}x compression)"
+            )
+
+            # Build compression metadata
+            compression_metadata = CompressionMetadata(
+                timestamp=datetime.now(timezone.utc),
+                query_index=compress_up_to_index,
+                compressed_summary=compressed_summary,
+                original_token_count=original_tokens,
+                compressed_token_count=compressed_tokens,
+                compression_ratio=compression_ratio,
+                model_used=self.model_id,
+                compression_prompt_version=self.prompt_builder.version,
+            )
+
+            return compression_metadata
+
+        except Exception as e:
+            logger.error(f"Error compressing conversation: {str(e)}", exc_info=True)
+            raise
+
+    def compress_and_save(
+        self,
+        conversation_id: str,
+        conversation: Dict[str, Any],
+        compress_up_to_index: int,
+    ) -> CompressionMetadata:
+        """
+        Compress conversation and save to database.
+
+        Args:
+            conversation_id: Conversation ID
+            conversation: Full conversation document
+            compress_up_to_index: Last query index to include
+
+        Returns:
+            CompressionMetadata
+
+        Raises:
+            ValueError: If conversation_service not provided or invalid index
+        """
+        if not self.conversation_service:
+            raise ValueError(
+                "conversation_service required for compress_and_save operation"
+            )
+
+        # Perform compression
+        metadata = self.compress_conversation(conversation, compress_up_to_index)
+
+        # Save to database
+        self.conversation_service.update_compression_metadata(
+            conversation_id, metadata.to_dict()
+        )
+
+        logger.info(f"Compression metadata saved to database for {conversation_id}")
+
+        return metadata
+
+    def get_compressed_context(
+        self, conversation: Dict[str, Any]
+    ) -> tuple[Optional[str], List[Dict[str, Any]]]:
+        """
+        Get compressed summary + recent uncompressed messages.
+
+        Args:
+            conversation: Full conversation document
+
+        Returns:
+            (compressed_summary, recent_messages)
+        """
+        try:
+            compression_metadata = conversation.get("compression_metadata", {})
+
+            if not compression_metadata.get("is_compressed"):
+                logger.debug("No compression metadata found - using full history")
+                queries = conversation.get("queries", [])
+                if queries is None:
+                    logger.error("Conversation queries is None - returning empty list")
+                    return None, []
+                return None, queries
+
+            compression_points = compression_metadata.get("compression_points", [])
+
+            if not compression_points:
+                logger.debug("No compression points found - using full history")
+                queries = conversation.get("queries", [])
+                if queries is None:
+                    logger.error("Conversation queries is None - returning empty list")
+                    return None, []
+                return None, queries
+
+            # Get the most recent compression point
+            latest_compression = compression_points[-1]
+            compressed_summary = latest_compression.get("compressed_summary")
+            last_compressed_index = latest_compression.get("query_index")
+            compressed_tokens = latest_compression.get("compressed_token_count", 0)
+            original_tokens = latest_compression.get("original_token_count", 0)
+
+            # Get only messages after compression point
+            queries = conversation.get("queries", [])
+            total_queries = len(queries)
+            recent_queries = queries[last_compressed_index + 1 :]
+
+            logger.info(
+                f"Using compressed context: summary ({compressed_tokens} tokens, "
+                f"compressed from {original_tokens}) + {len(recent_queries)} recent messages "
+                f"(messages {last_compressed_index + 1}-{total_queries - 1})"
+            )
+
+            return compressed_summary, recent_queries
+
+        except Exception as e:
+            logger.error(
+                f"Error getting compressed context: {str(e)}", exc_info=True
+            )
+            queries = conversation.get("queries", [])
+            if queries is None:
+                return None, []
+            return None, queries
+
+    def _extract_summary(self, llm_response: str) -> str:
+        """
+        Extract clean summary from LLM response.
+
+        Args:
+            llm_response: Raw LLM response
+
+        Returns:
+            Cleaned summary text
+        """
+        try:
+            # Try to extract content within <summary> tags
+            summary_match = re.search(
+                r"<summary>(.*?)</summary>", llm_response, re.DOTALL
+            )
+
+            if summary_match:
+                summary = summary_match.group(1).strip()
+            else:
+                # If no summary tags, remove analysis tags and use the rest
+                summary = re.sub(
+                    r"<analysis>.*?</analysis>", "", llm_response, flags=re.DOTALL
+                ).strip()
+
+            return summary
+
+        except Exception as e:
+            logger.warning(f"Error extracting summary: {str(e)}, using full response")
+            return llm_response
+
+    def _log_tool_call_stats(self, queries: List[Dict[str, Any]]) -> None:
+        """Log statistics about tool calls in queries."""
+        total_tool_calls = 0
+        total_tool_result_chars = 0
+        tool_call_breakdown = {}
+
+        for q in queries:
+            for tc in q.get("tool_calls", []):
+                total_tool_calls += 1
+                tool_name = tc.get("tool_name", "unknown")
+                action_name = tc.get("action_name", "unknown")
+                key = f"{tool_name}.{action_name}"
+                tool_call_breakdown[key] = tool_call_breakdown.get(key, 0) + 1
+
+                # Track total tool result size
+                result = tc.get("result", "")
+                if result:
+                    total_tool_result_chars += len(str(result))
+
+        if total_tool_calls > 0:
+            tool_breakdown_str = ", ".join(
+                f"{tool}({count})"
+                for tool, count in sorted(tool_call_breakdown.items())
+            )
+            tool_result_kb = total_tool_result_chars / 1024
+            logger.info(
+                f"Tool call breakdown: {tool_breakdown_str} "
+                f"(total result size: {tool_result_kb:.1f} KB, {total_tool_result_chars:,} chars)"
+            )

application/api/answer/services/compression/threshold_checker.py

@@ -0,0 +1,103 @@
+"""Compression threshold checking logic."""
+
+import logging
+from typing import Any, Dict
+
+from application.core.model_utils import get_token_limit
+from application.core.settings import settings
+from application.api.answer.services.compression.token_counter import TokenCounter
+
+logger = logging.getLogger(__name__)
+
+
+class CompressionThresholdChecker:
+    """Determines if compression is needed based on token thresholds."""
+
+    def __init__(self, threshold_percentage: float = None):
+        """
+        Initialize threshold checker.
+
+        Args:
+            threshold_percentage: Percentage of context to use as threshold
+                                 (defaults to settings.COMPRESSION_THRESHOLD_PERCENTAGE)
+        """
+        self.threshold_percentage = (
+            threshold_percentage or settings.COMPRESSION_THRESHOLD_PERCENTAGE
+        )
+
+    def should_compress(
+        self,
+        conversation: Dict[str, Any],
+        model_id: str,
+        current_query_tokens: int = 500,
+    ) -> bool:
+        """
+        Determine if compression is needed.
+
+        Args:
+            conversation: Full conversation document
+            model_id: Target model for this request
+            current_query_tokens: Estimated tokens for current query
+
+        Returns:
+            True if tokens >= threshold% of context window
+        """
+        try:
+            # Calculate total tokens in conversation
+            total_tokens = TokenCounter.count_conversation_tokens(conversation)
+            total_tokens += current_query_tokens
+
+            # Get context window limit for model
+            context_limit = get_token_limit(model_id)
+
+            # Calculate threshold
+            threshold = int(context_limit * self.threshold_percentage)
+
+            compression_needed = total_tokens >= threshold
+            percentage_used = (total_tokens / context_limit) * 100
+
+            if compression_needed:
+                logger.warning(
+                    f"COMPRESSION TRIGGERED: {total_tokens} tokens / {context_limit} limit "
+                    f"({percentage_used:.1f}% used, threshold: {self.threshold_percentage * 100:.0f}%)"
+                )
+            else:
+                logger.info(
+                    f"Compression check: {total_tokens}/{context_limit} tokens "
+                    f"({percentage_used:.1f}% used, threshold: {self.threshold_percentage * 100:.0f}%) - No compression needed"
+                )
+
+            return compression_needed
+
+        except Exception as e:
+            logger.error(f"Error checking compression need: {str(e)}", exc_info=True)
+            return False
+
+    def check_message_tokens(self, messages: list, model_id: str) -> bool:
+        """
+        Check if message list exceeds threshold.
+
+        Args:
+            messages: List of message dicts
+            model_id: Target model
+
+        Returns:
+            True if at or above threshold
+        """
+        try:
+            current_tokens = TokenCounter.count_message_tokens(messages)
+            context_limit = get_token_limit(model_id)
+            threshold = int(context_limit * self.threshold_percentage)
+
+            if current_tokens >= threshold:
+                logger.warning(
+                    f"Message context limit approaching: {current_tokens}/{context_limit} tokens "
+                    f"({(current_tokens/context_limit)*100:.1f}%)"
+                )
+                return True
+
+            return False
+
+        except Exception as e:
+            logger.error(f"Error checking message tokens: {str(e)}", exc_info=True)
+            return False

application/api/answer/services/compression/token_counter.py

@@ -0,0 +1,103 @@
+"""Token counting utilities for compression."""
+
+import logging
+from typing import Any, Dict, List
+
+from application.utils import num_tokens_from_string
+from application.core.settings import settings
+
+logger = logging.getLogger(__name__)
+
+
+class TokenCounter:
+    """Centralized token counting for conversations and messages."""
+
+    @staticmethod
+    def count_message_tokens(messages: List[Dict]) -> int:
+        """
+        Calculate total tokens in a list of messages.
+
+        Args:
+            messages: List of message dicts with 'content' field
+
+        Returns:
+            Total token count
+        """
+        total_tokens = 0
+        for message in messages:
+            content = message.get("content", "")
+            if isinstance(content, str):
+                total_tokens += num_tokens_from_string(content)
+            elif isinstance(content, list):
+                # Handle structured content (tool calls, etc.)
+                for item in content:
+                    if isinstance(item, dict):
+                        total_tokens += num_tokens_from_string(str(item))
+        return total_tokens
+
+    @staticmethod
+    def count_query_tokens(
+        queries: List[Dict[str, Any]], include_tool_calls: bool = True
+    ) -> int:
+        """
+        Count tokens across multiple query objects.
+
+        Args:
+            queries: List of query objects from conversation
+            include_tool_calls: Whether to count tool call tokens
+
+        Returns:
+            Total token count
+        """
+        total_tokens = 0
+
+        for query in queries:
+            # Count prompt and response tokens
+            if "prompt" in query:
+                total_tokens += num_tokens_from_string(query["prompt"])
+            if "response" in query:
+                total_tokens += num_tokens_from_string(query["response"])
+            if "thought" in query:
+                total_tokens += num_tokens_from_string(query.get("thought", ""))
+
+            # Count tool call tokens
+            if include_tool_calls and "tool_calls" in query:
+                for tool_call in query["tool_calls"]:
+                    tool_call_string = (
+                        f"Tool: {tool_call.get('tool_name')} | "
+                        f"Action: {tool_call.get('action_name')} | "
+                        f"Args: {tool_call.get('arguments')} | "
+                        f"Response: {tool_call.get('result')}"
+                    )
+                    total_tokens += num_tokens_from_string(tool_call_string)
+
+        return total_tokens
+
+    @staticmethod
+    def count_conversation_tokens(
+        conversation: Dict[str, Any], include_system_prompt: bool = False
+    ) -> int:
+        """
+        Calculate total tokens in a conversation.
+
+        Args:
+            conversation: Conversation document
+            include_system_prompt: Whether to include system prompt in count
+
+        Returns:
+            Total token count
+        """
+        try:
+            queries = conversation.get("queries", [])
+            total_tokens = TokenCounter.count_query_tokens(queries)
+
+            # Add system prompt tokens if requested
+            if include_system_prompt:
+                # Rough estimate for system prompt
+                total_tokens += settings.RESERVED_TOKENS.get("system_prompt", 500)
+
+            return total_tokens
+
+        except Exception as e:
+            logger.error(f"Error calculating conversation tokens: {str(e)}")
+            return 0

application/api/answer/services/compression/types.py

@@ -0,0 +1,83 @@
+"""Type definitions for compression module."""
+
+from dataclasses import dataclass, field
+from datetime import datetime
+from typing import Any, Dict, List, Optional
+
+
+@dataclass
+class CompressionMetadata:
+    """Metadata about a compression operation."""
+
+    timestamp: datetime
+    query_index: int
+    compressed_summary: str
+    original_token_count: int
+    compressed_token_count: int
+    compression_ratio: float
+    model_used: str
+    compression_prompt_version: str
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for DB storage."""
+        return {
+            "timestamp": self.timestamp,
+            "query_index": self.query_index,
+            "compressed_summary": self.compressed_summary,
+            "original_token_count": self.original_token_count,
+            "compressed_token_count": self.compressed_token_count,
+            "compression_ratio": self.compression_ratio,
+            "model_used": self.model_used,
+            "compression_prompt_version": self.compression_prompt_version,
+        }
+
+
+@dataclass
+class CompressionResult:
+    """Result of a compression operation."""
+
+    success: bool
+    compressed_summary: Optional[str] = None
+    recent_queries: List[Dict[str, Any]] = field(default_factory=list)
+    metadata: Optional[CompressionMetadata] = None
+    error: Optional[str] = None
+    compression_performed: bool = False
+
+    @classmethod
+    def success_with_compression(
+        cls, summary: str, queries: List[Dict], metadata: CompressionMetadata
+    ) -> "CompressionResult":
+        """Create a successful result with compression."""
+        return cls(
+            success=True,
+            compressed_summary=summary,
+            recent_queries=queries,
+            metadata=metadata,
+            compression_performed=True,
+        )
+
+    @classmethod
+    def success_no_compression(cls, queries: List[Dict]) -> "CompressionResult":
+        """Create a successful result without compression needed."""
+        return cls(
+            success=True,
+            recent_queries=queries,
+            compression_performed=False,
+        )
+
+    @classmethod
+    def failure(cls, error: str) -> "CompressionResult":
+        """Create a failure result."""
+        return cls(success=False, error=error, compression_performed=False)
+
+    def as_history(self) -> List[Dict[str, str]]:
+        """
+        Convert recent queries to history format.
+
+        Returns:
+            List of prompt/response dicts
+        """
+        return [
+            {"prompt": q["prompt"], "response": q["response"]}
+            for q in self.recent_queries
+        ]

application/api/answer/services/conversation_service.py

@@ -52,7 +52,7 @@ class ConversationService:
         sources: List[Dict[str, Any]],
         tool_calls: List[Dict[str, Any]],
         llm: Any,
-        gpt_model: str,
+        model_id: str,
         decoded_token: Dict[str, Any],
         index: Optional[int] = None,
         api_key: Optional[str] = None,
@@ -66,7 +66,7 @@ class ConversationService:
         if not user_id:
             raise ValueError("User ID not found in token")
         current_time = datetime.now(timezone.utc)
-        
+
         # clean up in sources array such that we save max 1k characters for text part
         for source in sources:
             if "text" in source and isinstance(source["text"], str):
@@ -90,6 +90,7 @@ class ConversationService:
                         f"queries.{index}.tool_calls": tool_calls,
                         f"queries.{index}.timestamp": current_time,
                         f"queries.{index}.attachments": attachment_ids,
+                        f"queries.{index}.model_id": model_id,
                     }
                 },
             )
@@ -120,6 +121,7 @@ class ConversationService:
                             "tool_calls": tool_calls,
                             "timestamp": current_time,
                             "attachments": attachment_ids,
+                            "model_id": model_id,
                         }
                     }
                 },
@@ -133,10 +135,9 @@ class ConversationService:
 
             messages_summary = [
                 {
-                    "role": "assistant",
-                    "content": "Summarise following conversation in no more than 3 "
-                    "words, respond ONLY with the summary, use the same "
-                    "language as the user query",
+                    "role": "system",
+                    "content": "You are a helpful assistant that creates concise conversation titles. "
+                    "Summarize conversations in 3 words or less using the same language as the user.",
                 },
                 {
                     "role": "user",
@@ -147,7 +148,7 @@ class ConversationService:
             ]
 
             completion = llm.gen(
-                model=gpt_model, messages=messages_summary, max_tokens=30
+                model=model_id, messages=messages_summary, max_tokens=30
             )
 
             conversation_data = {
@@ -163,6 +164,7 @@ class ConversationService:
                         "tool_calls": tool_calls,
                         "timestamp": current_time,
                         "attachments": attachment_ids,
+                        "model_id": model_id,
                     }
                 ],
             }
@@ -178,3 +180,103 @@ class ConversationService:
                     conversation_data["api_key"] = agent["key"]
             result = self.conversations_collection.insert_one(conversation_data)
             return str(result.inserted_id)
+
+    def update_compression_metadata(
+        self, conversation_id: str, compression_metadata: Dict[str, Any]
+    ) -> None:
+        """
+        Update conversation with compression metadata.
+
+        Uses $push with $slice to keep only the most recent compression points,
+        preventing unbounded array growth. Since each compression incorporates
+        previous compressions, older points become redundant.
+
+        Args:
+            conversation_id: Conversation ID
+            compression_metadata: Compression point data
+        """
+        try:
+            self.conversations_collection.update_one(
+                {"_id": ObjectId(conversation_id)},
+                {
+                    "$set": {
+                        "compression_metadata.is_compressed": True,
+                        "compression_metadata.last_compression_at": compression_metadata.get(
+                            "timestamp"
+                        ),
+                    },
+                    "$push": {
+                        "compression_metadata.compression_points": {
+                            "$each": [compression_metadata],
+                            "$slice": -settings.COMPRESSION_MAX_HISTORY_POINTS,
+                        }
+                    },
+                },
+            )
+            logger.info(
+                f"Updated compression metadata for conversation {conversation_id}"
+            )
+        except Exception as e:
+            logger.error(
+                f"Error updating compression metadata: {str(e)}", exc_info=True
+            )
+            raise
+
+    def append_compression_message(
+        self, conversation_id: str, compression_metadata: Dict[str, Any]
+    ) -> None:
+        """
+        Append a synthetic compression summary entry into the conversation history.
+        This makes the summary visible in the DB alongside normal queries.
+        """
+        try:
+            summary = compression_metadata.get("compressed_summary", "")
+            if not summary:
+                return
+            timestamp = compression_metadata.get("timestamp", datetime.now(timezone.utc))
+
+            self.conversations_collection.update_one(
+                {"_id": ObjectId(conversation_id)},
+                {
+                    "$push": {
+                        "queries": {
+                            "prompt": "[Context Compression Summary]",
+                            "response": summary,
+                            "thought": "",
+                            "sources": [],
+                            "tool_calls": [],
+                            "timestamp": timestamp,
+                            "attachments": [],
+                            "model_id": compression_metadata.get("model_used"),
+                        }
+                    }
+                },
+            )
+            logger.info(f"Appended compression summary to conversation {conversation_id}")
+        except Exception as e:
+            logger.error(
+                f"Error appending compression summary: {str(e)}", exc_info=True
+            )
+
+    def get_compression_metadata(
+        self, conversation_id: str
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Get compression metadata for a conversation.
+
+        Args:
+            conversation_id: Conversation ID
+
+        Returns:
+            Compression metadata dict or None
+        """
+        try:
+            conversation = self.conversations_collection.find_one(
+                {"_id": ObjectId(conversation_id)}, {"compression_metadata": 1}
+            )
+            return conversation.get("compression_metadata") if conversation else None
+        except Exception as e:
+            logger.error(
+                f"Error getting compression metadata: {str(e)}", exc_info=True
+            )
+            return None

application/api/answer/services/prompt_renderer.py

@@ -0,0 +1,97 @@
+import logging
+from typing import Any, Dict, Optional
+
+from application.templates.namespaces import NamespaceManager
+
+from application.templates.template_engine import TemplateEngine, TemplateRenderError
+
+logger = logging.getLogger(__name__)
+
+
+class PromptRenderer:
+    """Service for rendering prompts with dynamic context using namespaces"""
+
+    def __init__(self):
+        self.template_engine = TemplateEngine()
+        self.namespace_manager = NamespaceManager()
+
+    def render_prompt(
+        self,
+        prompt_content: str,
+        user_id: Optional[str] = None,
+        request_id: Optional[str] = None,
+        passthrough_data: Optional[Dict[str, Any]] = None,
+        docs: Optional[list] = None,
+        docs_together: Optional[str] = None,
+        tools_data: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ) -> str:
+        """
+        Render prompt with full context from all namespaces.
+
+        Args:
+            prompt_content: Raw prompt template string
+            user_id: Current user identifier
+            request_id: Unique request identifier
+            passthrough_data: Parameters from web request
+            docs: RAG retrieved documents
+            docs_together: Concatenated document content
+            tools_data: Pre-fetched tool results organized by tool name
+            **kwargs: Additional parameters for namespace builders
+
+        Returns:
+            Rendered prompt string with all variables substituted
+
+        Raises:
+            TemplateRenderError: If template rendering fails
+        """
+        if not prompt_content:
+            return ""
+
+        uses_template = self._uses_template_syntax(prompt_content)
+
+        if not uses_template:
+            return self._apply_legacy_substitutions(prompt_content, docs_together)
+
+        try:
+            context = self.namespace_manager.build_context(
+                user_id=user_id,
+                request_id=request_id,
+                passthrough_data=passthrough_data,
+                docs=docs,
+                docs_together=docs_together,
+                tools_data=tools_data,
+                **kwargs,
+            )
+
+            return self.template_engine.render(prompt_content, context)
+        except TemplateRenderError:
+            raise
+        except Exception as e:
+            error_msg = f"Prompt rendering failed: {str(e)}"
+            logger.error(error_msg)
+            raise TemplateRenderError(error_msg) from e
+
+    def _uses_template_syntax(self, prompt_content: str) -> bool:
+        """Check if prompt uses Jinja2 template syntax"""
+        return "{{" in prompt_content and "}}" in prompt_content
+
+    def _apply_legacy_substitutions(
+        self, prompt_content: str, docs_together: Optional[str] = None
+    ) -> str:
+        """
+        Apply backward-compatible substitutions for old prompt format.
+
+        Handles legacy {summaries} and {query} placeholders during transition period.
+        """
+        if docs_together:
+            prompt_content = prompt_content.replace("{summaries}", docs_together)
+        return prompt_content
+
+    def validate_template(self, prompt_content: str) -> bool:
+        """Validate prompt template syntax"""
+        return self.template_engine.validate_template(prompt_content)
+
+    def extract_variables(self, prompt_content: str) -> set[str]:
+        """Extract all variable names from prompt template"""
+        return self.template_engine.extract_variables(prompt_content)

application/api/answer/services/stream_processor.py

@@ -3,18 +3,30 @@ import json
 import logging
 import os
 from pathlib import Path
-from typing import Any, Dict, Optional
+from typing import Any, Dict, Optional, Set
 
 from bson.dbref import DBRef
 
 from bson.objectid import ObjectId
 
 from application.agents.agent_creator import AgentCreator
+from application.api.answer.services.compression import CompressionOrchestrator
+from application.api.answer.services.compression.token_counter import TokenCounter
 from application.api.answer.services.conversation_service import ConversationService
+from application.api.answer.services.prompt_renderer import PromptRenderer
+from application.core.model_utils import (
+    get_api_key_for_provider,
+    get_default_model_id,
+    get_provider_from_model_id,
+    validate_model_id,
+)
 from application.core.mongo_db import MongoDB
 from application.core.settings import settings
 from application.retriever.retriever_creator import RetrieverCreator
-from application.utils import get_gpt_model, limit_chat_history
+from application.utils import (
+    calculate_doc_token_budget,
+    limit_chat_history,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -73,19 +85,28 @@ class StreamProcessor:
         self.all_sources = []
         self.attachments = []
         self.history = []
+        self.retrieved_docs = []
         self.agent_config = {}
         self.retriever_config = {}
         self.is_shared_usage = False
         self.shared_token = None
-        self.gpt_model = get_gpt_model()
+        self.model_id: Optional[str] = None
         self.conversation_service = ConversationService()
+        self.compression_orchestrator = CompressionOrchestrator(
+            self.conversation_service
+        )
+        self.prompt_renderer = PromptRenderer()
+        self._prompt_content: Optional[str] = None
+        self._required_tool_actions: Optional[Dict[str, Set[Optional[str]]]] = None
+        self.compressed_summary: Optional[str] = None
+        self.compressed_summary_tokens: int = 0
 
     def initialize(self):
         """Initialize all required components for processing"""
         self._configure_agent()
+        self._validate_and_set_model()
         self._configure_source()
         self._configure_retriever()
-        self._configure_agent()
         self._load_conversation_history()
         self._process_attachments()
 
@@ -97,14 +118,71 @@ class StreamProcessor:
             )
             if not conversation:
                 raise ValueError("Conversation not found or unauthorized")
+
+            # Check if compression is enabled and needed
+            if settings.ENABLE_CONVERSATION_COMPRESSION:
+                self._handle_compression(conversation)
+            else:
+                # Original behavior - load all history
+                self.history = [
+                    {"prompt": query["prompt"], "response": query["response"]}
+                    for query in conversation.get("queries", [])
+                ]
+        else:
+            self.history = limit_chat_history(
+                json.loads(self.data.get("history", "[]")), model_id=self.model_id
+            )
+
+    def _handle_compression(self, conversation: Dict[str, Any]):
+        """
+        Handle conversation compression logic using orchestrator.
+
+        Args:
+            conversation: Full conversation document
+        """
+        try:
+            # Use orchestrator to handle all compression logic
+            result = self.compression_orchestrator.compress_if_needed(
+                conversation_id=self.conversation_id,
+                user_id=self.initial_user_id,
+                model_id=self.model_id,
+                decoded_token=self.decoded_token,
+            )
+
+            if not result.success:
+                logger.error(
+                    f"Compression failed: {result.error}, using full history"
+                )
+                self.history = [
+                    {"prompt": query["prompt"], "response": query["response"]}
+                    for query in conversation.get("queries", [])
+                ]
+                return
+
+            # Set compressed summary if compression was performed
+            if result.compression_performed and result.compressed_summary:
+                self.compressed_summary = result.compressed_summary
+                self.compressed_summary_tokens = TokenCounter.count_message_tokens(
+                    [{"content": result.compressed_summary}]
+                )
+                logger.info(
+                    f"Using compressed summary ({self.compressed_summary_tokens} tokens) "
+                    f"+ {len(result.recent_queries)} recent messages"
+                )
+
+            # Build history from recent queries
+            self.history = result.as_history()
+
+        except Exception as e:
+            logger.error(
+                f"Error handling compression, falling back to standard history: {str(e)}",
+                exc_info=True,
+            )
+            # Fallback to original behavior
             self.history = [
                 {"prompt": query["prompt"], "response": query["response"]}
                 for query in conversation.get("queries", [])
             ]
-        else:
-            self.history = limit_chat_history(
-                json.loads(self.data.get("history", "[]")), gpt_model=self.gpt_model
-            )
 
     def _process_attachments(self):
         """Process any attachments in the request"""
@@ -134,6 +212,30 @@ class StreamProcessor:
                 )
         return attachments
 
+    def _validate_and_set_model(self):
+        """Validate and set model_id from request"""
+        from application.core.model_settings import ModelRegistry
+
+        requested_model = self.data.get("model_id")
+
+        if requested_model:
+            if not validate_model_id(requested_model):
+                registry = ModelRegistry.get_instance()
+                available_models = [m.id for m in registry.get_enabled_models()]
+                raise ValueError(
+                    f"Invalid model_id '{requested_model}'. "
+                    f"Available models: {', '.join(available_models[:5])}"
+                    + (f" and {len(available_models) - 5} more" if len(available_models) > 5 else "")
+                )
+            self.model_id = requested_model
+        else:
+            # Check if agent has a default model configured
+            agent_default_model = self.agent_config.get("default_model_id", "")
+            if agent_default_model and validate_model_id(agent_default_model):
+                self.model_id = agent_default_model
+            else:
+                self.model_id = get_default_model_id()
+
     def _get_agent_key(self, agent_id: Optional[str], user_id: Optional[str]) -> tuple:
         """Get API key for agent with access control"""
         if not agent_id:
@@ -205,6 +307,10 @@ class StreamProcessor:
             data["sources"] = sources_list
         else:
             data["sources"] = []
+
+        # Preserve model configuration from agent
+        data["default_model_id"] = data.get("default_model_id", "")
+
         return data
 
     def _configure_source(self):
@@ -257,6 +363,7 @@ class StreamProcessor:
                     "agent_type": data_key.get("agent_type", settings.AGENT_NAME),
                     "user_api_key": api_key,
                     "json_schema": data_key.get("json_schema"),
+                    "default_model_id": data_key.get("default_model_id", ""),
                 }
             )
             self.initial_user_id = data_key.get("user")
@@ -281,6 +388,7 @@ class StreamProcessor:
                     "agent_type": data_key.get("agent_type", settings.AGENT_NAME),
                     "user_api_key": self.agent_key,
                     "json_schema": data_key.get("json_schema"),
+                    "default_model_id": data_key.get("default_model_id", ""),
                 }
             )
             self.decoded_token = (
@@ -307,47 +415,341 @@ class StreamProcessor:
                     "agent_type": settings.AGENT_NAME,
                     "user_api_key": None,
                     "json_schema": None,
+                    "default_model_id": "",
                 }
             )
 
     def _configure_retriever(self):
-        """Configure the retriever based on request data"""
+        history_token_limit = int(self.data.get("token_limit", 2000))
+        doc_token_limit = calculate_doc_token_budget(
+            model_id=self.model_id, history_token_limit=history_token_limit
+        )
+
         self.retriever_config = {
             "retriever_name": self.data.get("retriever", "classic"),
             "chunks": int(self.data.get("chunks", 2)),
-            "token_limit": self.data.get("token_limit", settings.DEFAULT_MAX_HISTORY),
+            "doc_token_limit": doc_token_limit,
+            "history_token_limit": history_token_limit,
         }
 
         api_key = self.data.get("api_key") or self.agent_key
         if not api_key and "isNoneDoc" in self.data and self.data["isNoneDoc"]:
             self.retriever_config["chunks"] = 0
 
-    def create_agent(self):
-        """Create and return the configured agent"""
-        return AgentCreator.create_agent(
-            self.agent_config["agent_type"],
-            endpoint="stream",
-            llm_name=settings.LLM_PROVIDER,
-            gpt_model=self.gpt_model,
-            api_key=settings.API_KEY,
-            user_api_key=self.agent_config["user_api_key"],
-            prompt=get_prompt(self.agent_config["prompt_id"], self.prompts_collection),
-            chat_history=self.history,
-            decoded_token=self.decoded_token,
-            attachments=self.attachments,
-            json_schema=self.agent_config.get("json_schema"),
-        )
-
     def create_retriever(self):
-        """Create and return the configured retriever"""
         return RetrieverCreator.create_retriever(
             self.retriever_config["retriever_name"],
             source=self.source,
             chat_history=self.history,
             prompt=get_prompt(self.agent_config["prompt_id"], self.prompts_collection),
             chunks=self.retriever_config["chunks"],
-            token_limit=self.retriever_config["token_limit"],
-            gpt_model=self.gpt_model,
+            doc_token_limit=self.retriever_config.get("doc_token_limit", 50000),
+            model_id=self.model_id,
             user_api_key=self.agent_config["user_api_key"],
             decoded_token=self.decoded_token,
         )
+
+    def pre_fetch_docs(self, question: str) -> tuple[Optional[str], Optional[list]]:
+        """Pre-fetch documents for template rendering before agent creation"""
+        if self.data.get("isNoneDoc", False):
+            logger.info("Pre-fetch skipped: isNoneDoc=True")
+            return None, None
+        try:
+            retriever = self.create_retriever()
+            logger.info(
+                f"Pre-fetching docs with chunks={retriever.chunks}, doc_token_limit={retriever.doc_token_limit}"
+            )
+            docs = retriever.search(question)
+            logger.info(f"Pre-fetch retrieved {len(docs) if docs else 0} documents")
+
+            if not docs:
+                logger.info("Pre-fetch: No documents returned from search")
+                return None, None
+            self.retrieved_docs = docs
+
+            docs_with_filenames = []
+            for doc in docs:
+                filename = doc.get("filename") or doc.get("title") or doc.get("source")
+                if filename:
+                    chunk_header = str(filename)
+                    docs_with_filenames.append(f"{chunk_header}\n{doc['text']}")
+                else:
+                    docs_with_filenames.append(doc["text"])
+            docs_together = "\n\n".join(docs_with_filenames)
+
+            logger.info(f"Pre-fetch docs_together size: {len(docs_together)} chars")
+
+            return docs_together, docs
+        except Exception as e:
+            logger.error(f"Failed to pre-fetch docs: {str(e)}", exc_info=True)
+            return None, None
+
+    def pre_fetch_tools(self) -> Optional[Dict[str, Any]]:
+        """Pre-fetch tool data for template rendering before agent creation
+
+        Can be controlled via:
+        1. Global setting: ENABLE_TOOL_PREFETCH in .env
+        2. Per-request: disable_tool_prefetch in request data
+        """
+        if not settings.ENABLE_TOOL_PREFETCH:
+            logger.info(
+                "Tool pre-fetching disabled globally via ENABLE_TOOL_PREFETCH setting"
+            )
+            return None
+
+        if self.data.get("disable_tool_prefetch", False):
+            logger.info("Tool pre-fetching disabled for this request")
+            return None
+
+        required_tool_actions = self._get_required_tool_actions()
+        filtering_enabled = required_tool_actions is not None
+
+        try:
+            user_tools_collection = self.db["user_tools"]
+            user_id = self.initial_user_id or "local"
+
+            user_tools = list(
+                user_tools_collection.find({"user": user_id, "status": True})
+            )
+
+            if not user_tools:
+                return None
+
+            tools_data = {}
+
+            for tool_doc in user_tools:
+                tool_name = tool_doc.get("name")
+                tool_id = str(tool_doc.get("_id"))
+
+                if filtering_enabled:
+                    required_actions_by_name = required_tool_actions.get(
+                        tool_name, set()
+                    )
+                    required_actions_by_id = required_tool_actions.get(tool_id, set())
+
+                    required_actions = required_actions_by_name | required_actions_by_id
+
+                    if not required_actions:
+                        continue
+                else:
+                    required_actions = None
+
+                tool_data = self._fetch_tool_data(tool_doc, required_actions)
+                if tool_data:
+                    tools_data[tool_name] = tool_data
+                    tools_data[tool_id] = tool_data
+
+            return tools_data if tools_data else None
+        except Exception as e:
+            logger.warning(f"Failed to pre-fetch tools: {type(e).__name__}")
+            return None
+
+    def _fetch_tool_data(
+        self,
+        tool_doc: Dict[str, Any],
+        required_actions: Optional[Set[Optional[str]]],
+    ) -> Optional[Dict[str, Any]]:
+        """Fetch and execute tool actions with saved parameters"""
+        try:
+            from application.agents.tools.tool_manager import ToolManager
+
+            tool_name = tool_doc.get("name")
+            tool_config = tool_doc.get("config", {}).copy()
+            tool_config["tool_id"] = str(tool_doc["_id"])
+
+            tool_manager = ToolManager(config={tool_name: tool_config})
+            user_id = self.initial_user_id or "local"
+            tool = tool_manager.load_tool(tool_name, tool_config, user_id=user_id)
+
+            if not tool:
+                logger.debug(f"Tool '{tool_name}' failed to load")
+                return None
+
+            tool_actions = tool.get_actions_metadata()
+            if not tool_actions:
+                logger.debug(f"Tool '{tool_name}' has no actions")
+                return None
+
+            saved_actions = tool_doc.get("actions", [])
+
+            include_all_actions = required_actions is None or (
+                required_actions and None in required_actions
+            )
+            allowed_actions: Set[str] = (
+                {action for action in required_actions if isinstance(action, str)}
+                if required_actions
+                else set()
+            )
+
+            action_results = {}
+            for action_meta in tool_actions:
+                action_name = action_meta.get("name")
+                if action_name is None:
+                    continue
+                if (
+                    not include_all_actions
+                    and allowed_actions
+                    and action_name not in allowed_actions
+                ):
+                    continue
+
+                try:
+                    saved_action = None
+                    for sa in saved_actions:
+                        if sa.get("name") == action_name:
+                            saved_action = sa
+                            break
+
+                    action_params = action_meta.get("parameters", {})
+                    properties = action_params.get("properties", {})
+
+                    kwargs = {}
+                    for param_name, param_spec in properties.items():
+                        if saved_action:
+                            saved_props = saved_action.get("parameters", {}).get(
+                                "properties", {}
+                            )
+                            if param_name in saved_props:
+                                param_value = saved_props[param_name].get("value")
+                                if param_value is not None:
+                                    kwargs[param_name] = param_value
+                                    continue
+
+                        if param_name in tool_config:
+                            kwargs[param_name] = tool_config[param_name]
+                        elif "default" in param_spec:
+                            kwargs[param_name] = param_spec["default"]
+
+                    result = tool.execute_action(action_name, **kwargs)
+                    action_results[action_name] = result
+                except Exception as e:
+                    logger.debug(
+                        f"Action '{action_name}' execution failed: {type(e).__name__}"
+                    )
+                    continue
+
+            return action_results if action_results else None
+
+        except Exception as e:
+            logger.debug(f"Tool pre-fetch failed for '{tool_name}': {type(e).__name__}")
+            return None
+
+    def _get_prompt_content(self) -> Optional[str]:
+        """Retrieve and cache the raw prompt content for the current agent configuration."""
+        if self._prompt_content is not None:
+            return self._prompt_content
+        prompt_id = (
+            self.agent_config.get("prompt_id")
+            if isinstance(self.agent_config, dict)
+            else None
+        )
+        if not prompt_id:
+            return None
+        try:
+            self._prompt_content = get_prompt(prompt_id, self.prompts_collection)
+        except ValueError as e:
+            logger.debug(f"Invalid prompt ID '{prompt_id}': {str(e)}")
+            self._prompt_content = None
+        except Exception as e:
+            logger.debug(f"Failed to fetch prompt '{prompt_id}': {type(e).__name__}")
+            self._prompt_content = None
+        return self._prompt_content
+
+    def _get_required_tool_actions(self) -> Optional[Dict[str, Set[Optional[str]]]]:
+        """Determine which tool actions are referenced in the prompt template"""
+        if self._required_tool_actions is not None:
+            return self._required_tool_actions
+
+        prompt_content = self._get_prompt_content()
+        if prompt_content is None:
+            return None
+
+        if "{{" not in prompt_content or "}}" not in prompt_content:
+            self._required_tool_actions = {}
+            return self._required_tool_actions
+
+        try:
+            from application.templates.template_engine import TemplateEngine
+
+            template_engine = TemplateEngine()
+            usages = template_engine.extract_tool_usages(prompt_content)
+            self._required_tool_actions = usages
+            return self._required_tool_actions
+        except Exception as e:
+            logger.debug(f"Failed to extract tool usages: {type(e).__name__}")
+            self._required_tool_actions = {}
+            return self._required_tool_actions
+
+    def _fetch_memory_tool_data(
+        self, tool_doc: Dict[str, Any]
+    ) -> Optional[Dict[str, Any]]:
+        """Fetch memory tool data for pre-injection into prompt"""
+        try:
+            tool_config = tool_doc.get("config", {}).copy()
+            tool_config["tool_id"] = str(tool_doc["_id"])
+
+            from application.agents.tools.memory import MemoryTool
+
+            memory_tool = MemoryTool(tool_config, self.initial_user_id)
+
+            root_view = memory_tool.execute_action("view", path="/")
+
+            if "Error:" in root_view or not root_view.strip():
+                return None
+
+            return {"root": root_view, "available": True}
+        except Exception as e:
+            logger.warning(f"Failed to fetch memory tool data: {str(e)}")
+            return None
+
+    def create_agent(
+        self,
+        docs_together: Optional[str] = None,
+        docs: Optional[list] = None,
+        tools_data: Optional[Dict[str, Any]] = None,
+    ):
+        """Create and return the configured agent with rendered prompt"""
+        raw_prompt = self._get_prompt_content()
+        if raw_prompt is None:
+            raw_prompt = get_prompt(
+                self.agent_config["prompt_id"], self.prompts_collection
+            )
+            self._prompt_content = raw_prompt
+
+        rendered_prompt = self.prompt_renderer.render_prompt(
+            prompt_content=raw_prompt,
+            user_id=self.initial_user_id,
+            request_id=self.data.get("request_id"),
+            passthrough_data=self.data.get("passthrough"),
+            docs=docs,
+            docs_together=docs_together,
+            tools_data=tools_data,
+        )
+
+        provider = (
+            get_provider_from_model_id(self.model_id)
+            if self.model_id
+            else settings.LLM_PROVIDER
+        )
+        system_api_key = get_api_key_for_provider(provider or settings.LLM_PROVIDER)
+
+        agent = AgentCreator.create_agent(
+            self.agent_config["agent_type"],
+            endpoint="stream",
+            llm_name=provider or settings.LLM_PROVIDER,
+            model_id=self.model_id,
+            api_key=system_api_key,
+            user_api_key=self.agent_config["user_api_key"],
+            prompt=rendered_prompt,
+            chat_history=self.history,
+            retrieved_docs=self.retrieved_docs,
+            decoded_token=self.decoded_token,
+            attachments=self.attachments,
+            json_schema=self.agent_config.get("json_schema"),
+            compressed_summary=self.compressed_summary,
+        )
+
+        agent.conversation_id = self.conversation_id
+        agent.initial_user_id = self.initial_user_id
+
+        return agent

application/api/user/agents/routes.py

@@ -10,7 +10,6 @@ from flask import current_app, jsonify, make_response, request
 from flask_restx import fields, Namespace, Resource
 
 from application.api import api
-from application.core.settings import settings
 from application.api.user.base import (
     agents_collection,
     db,
@@ -20,6 +19,7 @@ from application.api.user.base import (
     storage,
     users_collection,
 )
+from application.core.settings import settings
 from application.utils import (
     check_required_fields,
     generate_image_url,
@@ -76,9 +76,13 @@ class GetAgent(Resource):
                 "status": agent.get("status", ""),
                 "json_schema": agent.get("json_schema"),
                 "limited_token_mode": agent.get("limited_token_mode", False),
-                "token_limit": agent.get("token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]),
+                "token_limit": agent.get(
+                    "token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]
+                ),
                 "limited_request_mode": agent.get("limited_request_mode", False),
-                "request_limit": agent.get("request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]),
+                "request_limit": agent.get(
+                    "request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]
+                ),
                 "created_at": agent.get("createdAt", ""),
                 "updated_at": agent.get("updatedAt", ""),
                 "last_used_at": agent.get("lastUsedAt", ""),
@@ -91,6 +95,8 @@ class GetAgent(Resource):
                 "shared": agent.get("shared_publicly", False),
                 "shared_metadata": agent.get("shared_metadata", {}),
                 "shared_token": agent.get("shared_token", ""),
+                "models": agent.get("models", []),
+                "default_model_id": agent.get("default_model_id", ""),
             }
             return make_response(jsonify(data), 200)
         except Exception as e:
@@ -149,9 +155,13 @@ class GetAgents(Resource):
                     "status": agent.get("status", ""),
                     "json_schema": agent.get("json_schema"),
                     "limited_token_mode": agent.get("limited_token_mode", False),
-                    "token_limit": agent.get("token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]),
+                    "token_limit": agent.get(
+                        "token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]
+                    ),
                     "limited_request_mode": agent.get("limited_request_mode", False),
-                    "request_limit": agent.get("request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]),
+                    "request_limit": agent.get(
+                        "request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]
+                    ),
                     "created_at": agent.get("createdAt", ""),
                     "updated_at": agent.get("updatedAt", ""),
                     "last_used_at": agent.get("lastUsedAt", ""),
@@ -164,6 +174,8 @@ class GetAgents(Resource):
                     "shared": agent.get("shared_publicly", False),
                     "shared_metadata": agent.get("shared_metadata", {}),
                     "shared_token": agent.get("shared_token", ""),
+                    "models": agent.get("models", []),
+                    "default_model_id": agent.get("default_model_id", ""),
                 }
                 for agent in agents
                 if "source" in agent or "retriever" in agent
@@ -209,21 +221,27 @@ class CreateAgent(Resource):
                 description="JSON schema for enforcing structured output format",
             ),
             "limited_token_mode": fields.Boolean(
-                required=False,
-                description="Whether the agent is in limited token mode"
+                required=False, description="Whether the agent is in limited token mode"
             ),
             "token_limit": fields.Integer(
-                required=False,
-                description="Token limit for the agent in limited mode"
+                required=False, description="Token limit for the agent in limited mode"
             ),
             "limited_request_mode": fields.Boolean(
                 required=False,
-                description="Whether the agent is in limited request mode"
+                description="Whether the agent is in limited request mode",
             ),
             "request_limit": fields.Integer(
                 required=False,
-                description="Request limit for the agent in limited mode"
-            )
+                description="Request limit for the agent in limited mode",
+            ),
+            "models": fields.List(
+                fields.String,
+                required=False,
+                description="List of available model IDs for this agent",
+            ),
+            "default_model_id": fields.String(
+                required=False, description="Default model ID for this agent"
+            ),
         },
     )
 
@@ -252,6 +270,11 @@ class CreateAgent(Resource):
                     data["json_schema"] = json.loads(data["json_schema"])
                 except json.JSONDecodeError:
                     data["json_schema"] = None
+            if "models" in data:
+                try:
+                    data["models"] = json.loads(data["models"])
+                except json.JSONDecodeError:
+                    data["models"] = []
         print(f"Received data: {data}")
 
         # Validate JSON schema if provided
@@ -369,14 +392,32 @@ class CreateAgent(Resource):
                 "agent_type": data.get("agent_type", ""),
                 "status": data.get("status"),
                 "json_schema": data.get("json_schema"),
-                "limited_token_mode": data.get("limited_token_mode", False),
-                "token_limit": data.get("token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]),
-                "limited_request_mode": data.get("limited_request_mode", False),
-                "request_limit": data.get("request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]),
+                "limited_token_mode": (
+                    data.get("limited_token_mode") == "True"
+                    if isinstance(data.get("limited_token_mode"), str)
+                    else bool(data.get("limited_token_mode", False))
+                ),
+                "token_limit": int(
+                    data.get(
+                        "token_limit", settings.DEFAULT_AGENT_LIMITS["token_limit"]
+                    )
+                ),
+                "limited_request_mode": (
+                    data.get("limited_request_mode") == "True"
+                    if isinstance(data.get("limited_request_mode"), str)
+                    else bool(data.get("limited_request_mode", False))
+                ),
+                "request_limit": int(
+                    data.get(
+                        "request_limit", settings.DEFAULT_AGENT_LIMITS["request_limit"]
+                    )
+                ),
                 "createdAt": datetime.datetime.now(datetime.timezone.utc),
                 "updatedAt": datetime.datetime.now(datetime.timezone.utc),
                 "lastUsedAt": None,
                 "key": key,
+                "models": data.get("models", []),
+                "default_model_id": data.get("default_model_id", ""),
             }
             if new_agent["chunks"] == "":
                 new_agent["chunks"] = "2"
@@ -429,21 +470,27 @@ class UpdateAgent(Resource):
                 description="JSON schema for enforcing structured output format",
             ),
             "limited_token_mode": fields.Boolean(
-                required=False,
-                description="Whether the agent is in limited token mode"
+                required=False, description="Whether the agent is in limited token mode"
             ),
             "token_limit": fields.Integer(
-                required=False,
-                description="Token limit for the agent in limited mode"
+                required=False, description="Token limit for the agent in limited mode"
             ),
             "limited_request_mode": fields.Boolean(
                 require=False,
-                description="Whether the agent is in limited request mode"
+                description="Whether the agent is in limited request mode",
             ),
             "request_limit": fields.Integer(
                 required=False,
-                description="Request limit for the agent in limited mode"
-            )
+                description="Request limit for the agent in limited mode",
+            ),
+            "models": fields.List(
+                fields.String,
+                required=False,
+                description="List of available model IDs for this agent",
+            ),
+            "default_model_id": fields.String(
+                required=False, description="Default model ID for this agent"
+            ),
         },
     )
 
@@ -467,7 +514,7 @@ class UpdateAgent(Resource):
                 data = request.get_json()
             else:
                 data = request.form.to_dict()
-                json_fields = ["tools", "sources", "json_schema"]
+                json_fields = ["tools", "sources", "json_schema", "models"]
                 for field in json_fields:
                     if field in data and data[field]:
                         try:
@@ -534,7 +581,9 @@ class UpdateAgent(Resource):
             "limited_token_mode",
             "token_limit",
             "limited_request_mode",
-            "request_limit"
+            "request_limit",
+            "models",
+            "default_model_id",
         ]
 
         for field in allowed_fields:
@@ -652,8 +701,15 @@ class UpdateAgent(Resource):
                 else:
                     update_fields[field] = None
             elif field == "limited_token_mode":
-                is_mode_enabled = data.get("limited_token_mode", False)
-                if is_mode_enabled and data.get("token_limit") is None:
+                raw_value = data.get("limited_token_mode", False)
+                bool_value = (
+                    raw_value == "True"
+                    if isinstance(raw_value, str)
+                    else bool(raw_value)
+                )
+                update_fields[field] = bool_value
+
+                if bool_value and data.get("token_limit") is None:
                     return make_response(
                         jsonify(
                             {
@@ -664,8 +720,15 @@ class UpdateAgent(Resource):
                         400,
                     )
             elif field == "limited_request_mode":
-                is_mode_enabled = data.get("limited_request_mode", False)
-                if is_mode_enabled and data.get("request_limit") is None:
+                raw_value = data.get("limited_request_mode", False)
+                bool_value = (
+                    raw_value == "True"
+                    if isinstance(raw_value, str)
+                    else bool(raw_value)
+                )
+                update_fields[field] = bool_value
+
+                if bool_value and data.get("request_limit") is None:
                     return make_response(
                         jsonify(
                             {
@@ -677,7 +740,11 @@ class UpdateAgent(Resource):
                     )
             elif field == "token_limit":
                 token_limit = data.get("token_limit")
-                if token_limit is not None and not data.get("limited_token_mode"):
+                # Convert to int and store
+                update_fields[field] = int(token_limit) if token_limit else 0
+
+                # Validate consistency with mode
+                if update_fields[field] > 0 and not data.get("limited_token_mode"):
                     return make_response(
                         jsonify(
                             {
@@ -689,7 +756,9 @@ class UpdateAgent(Resource):
                     )
             elif field == "request_limit":
                 request_limit = data.get("request_limit")
-                if request_limit is not None and not data.get("limited_request_mode"):
+                update_fields[field] = int(request_limit) if request_limit else 0
+
+                if update_fields[field] > 0 and not data.get("limited_request_mode"):
                     return make_response(
                         jsonify(
                             {

application/api/user/attachments/routes.py

@@ -25,7 +25,7 @@ class StoreAttachment(Resource):
         api.model(
             "AttachmentModel",
             {
-                "file": fields.Raw(required=True, description="File to upload"),
+                "file": fields.Raw(required=True, description="File(s) to upload"),
                 "api_key": fields.String(
                     required=False, description="API key (optional)"
                 ),
@@ -33,18 +33,24 @@ class StoreAttachment(Resource):
         )
     )
     @api.doc(
-        description="Stores a single attachment without vectorization or training. Supports user or API key authentication."
+        description="Stores one or multiple attachments without vectorization or training. Supports user or API key authentication."
     )
     def post(self):
         decoded_token = getattr(request, "decoded_token", None)
         api_key = request.form.get("api_key") or request.args.get("api_key")
-        file = request.files.get("file")
-
-        if not file or file.filename == "":
+        
+        files = request.files.getlist("file")
+        if not files:
+            single_file = request.files.get("file")
+            if single_file:
+                files = [single_file]
+        
+        if not files or all(f.filename == "" for f in files):
             return make_response(
-                jsonify({"status": "error", "message": "Missing file"}),
+                jsonify({"status": "error", "message": "Missing file(s)"}),
                 400,
             )
+        
         user = None
         if decoded_token:
             user = safe_filename(decoded_token.get("sub"))
@@ -59,32 +65,74 @@ class StoreAttachment(Resource):
             return make_response(
                 jsonify({"success": False, "message": "Authentication required"}), 401
             )
+        
         try:
-            attachment_id = ObjectId()
-            original_filename = safe_filename(os.path.basename(file.filename))
-            relative_path = f"{settings.UPLOAD_FOLDER}/{user}/attachments/{str(attachment_id)}/{original_filename}"
+            tasks = []
+            errors = []
+            original_file_count = len(files)
+            
+            for idx, file in enumerate(files):
+                try:
+                    attachment_id = ObjectId()
+                    original_filename = safe_filename(os.path.basename(file.filename))
+                    relative_path = f"{settings.UPLOAD_FOLDER}/{user}/attachments/{str(attachment_id)}/{original_filename}"
 
-            metadata = storage.save_file(file, relative_path)
-
-            file_info = {
-                "filename": original_filename,
-                "attachment_id": str(attachment_id),
-                "path": relative_path,
-                "metadata": metadata,
-            }
-
-            task = store_attachment.delay(file_info, user)
-
-            return make_response(
-                jsonify(
-                    {
-                        "success": True,
-                        "task_id": task.id,
-                        "message": "File uploaded successfully. Processing started.",
+                    metadata = storage.save_file(file, relative_path)
+                    file_info = {
+                        "filename": original_filename,
+                        "attachment_id": str(attachment_id),
+                        "path": relative_path,
+                        "metadata": metadata,
                     }
-                ),
-                200,
-            )
+
+                    task = store_attachment.delay(file_info, user)
+                    tasks.append({
+                        "task_id": task.id,
+                        "filename": original_filename,
+                        "attachment_id": str(attachment_id),
+                    })
+                except Exception as file_err:
+                    current_app.logger.error(f"Error processing file {idx} ({file.filename}): {file_err}", exc_info=True)
+                    errors.append({
+                        "filename": file.filename,
+                        "error": str(file_err)
+                    })
+            
+            if not tasks:
+                error_msg = "No valid files to upload"
+                if errors:
+                    error_msg += f". Errors: {errors}"
+                return make_response(
+                    jsonify({"status": "error", "message": error_msg, "errors": errors}),
+                    400,
+                )
+            
+            if original_file_count == 1 and len(tasks) == 1:
+                current_app.logger.info("Returning single task_id response")
+                return make_response(
+                    jsonify(
+                        {
+                            "success": True,
+                            "task_id": tasks[0]["task_id"],
+                            "message": "File uploaded successfully. Processing started.",
+                        }
+                    ),
+                    200,
+                )
+            else:
+                response_data = {
+                    "success": True,
+                    "tasks": tasks,
+                    "message": f"{len(tasks)} file(s) uploaded successfully. Processing started.",
+                }
+                if errors:
+                    response_data["errors"] = errors
+                    response_data["message"] += f" {len(errors)} file(s) failed."
+                
+                return make_response(
+                    jsonify(response_data),
+                    200,
+                )
         except Exception as err:
             current_app.logger.error(f"Error storing attachment: {err}", exc_info=True)
             return make_response(jsonify({"success": False, "error": str(err)}), 400)

application/api/user/models/__init__.py

@@ -0,0 +1,3 @@
+from .routes import models_ns
+
+__all__ = ["models_ns"]

application/api/user/models/routes.py

@@ -0,0 +1,25 @@
+from flask import current_app, jsonify, make_response
+from flask_restx import Namespace, Resource
+
+from application.core.model_settings import ModelRegistry
+
+models_ns = Namespace("models", description="Available models", path="/api")
+
+
+@models_ns.route("/models")
+class ModelsListResource(Resource):
+    def get(self):
+        """Get list of available models with their capabilities."""
+        try:
+            registry = ModelRegistry.get_instance()
+            models = registry.get_enabled_models()
+
+            response = {
+                "models": [model.to_dict() for model in models],
+                "default_model_id": registry.default_model_id,
+                "count": len(models),
+            }
+        except Exception as err:
+            current_app.logger.error(f"Error fetching models: {err}", exc_info=True)
+            return make_response(jsonify({"success": False}), 500)
+        return make_response(jsonify(response), 200)

application/api/user/routes.py

@@ -10,6 +10,7 @@ from .agents import agents_ns, agents_sharing_ns, agents_webhooks_ns
 from .analytics import analytics_ns
 from .attachments import attachments_ns
 from .conversations import conversations_ns
+from .models import models_ns
 from .prompts import prompts_ns
 from .sharing import sharing_ns
 from .sources import sources_chunks_ns, sources_ns, sources_upload_ns
@@ -27,6 +28,9 @@ api.add_namespace(attachments_ns)
 # Conversations
 api.add_namespace(conversations_ns)
 
+# Models
+api.add_namespace(models_ns)
+
 # Agents (main, sharing, webhooks)
 api.add_namespace(agents_ns)
 api.add_namespace(agents_sharing_ns)

application/api/user/sharing/routes.py

@@ -13,7 +13,6 @@ from application.api.user.base import (
     agents_collection,
     attachments_collection,
     conversations_collection,
-    db,
     shared_conversations_collections,
 )
 from application.utils import check_required_fields
@@ -97,9 +96,7 @@ class ShareConversation(Resource):
                     api_uuid = pre_existing_api_document["key"]
                     pre_existing = shared_conversations_collections.find_one(
                         {
-                            "conversation_id": DBRef(
-                                "conversations", ObjectId(conversation_id)
-                            ),
+                            "conversation_id": ObjectId(conversation_id),
                             "isPromptable": is_promptable,
                             "first_n_queries": current_n_queries,
                             "user": user,
@@ -120,10 +117,7 @@ class ShareConversation(Resource):
                         shared_conversations_collections.insert_one(
                             {
                                 "uuid": explicit_binary,
-                                "conversation_id": {
-                                    "$ref": "conversations",
-                                    "$id": ObjectId(conversation_id),
-                                },
+                                "conversation_id": ObjectId(conversation_id),
                                 "isPromptable": is_promptable,
                                 "first_n_queries": current_n_queries,
                                 "user": user,
@@ -154,10 +148,7 @@ class ShareConversation(Resource):
                     shared_conversations_collections.insert_one(
                         {
                             "uuid": explicit_binary,
-                            "conversation_id": {
-                                "$ref": "conversations",
-                                "$id": ObjectId(conversation_id),
-                            },
+                            "conversation_id": ObjectId(conversation_id),
                             "isPromptable": is_promptable,
                             "first_n_queries": current_n_queries,
                             "user": user,
@@ -175,9 +166,7 @@ class ShareConversation(Resource):
                     )
             pre_existing = shared_conversations_collections.find_one(
                 {
-                    "conversation_id": DBRef(
-                        "conversations", ObjectId(conversation_id)
-                    ),
+                    "conversation_id": ObjectId(conversation_id),
                     "isPromptable": is_promptable,
                     "first_n_queries": current_n_queries,
                     "user": user,
@@ -197,10 +186,7 @@ class ShareConversation(Resource):
                 shared_conversations_collections.insert_one(
                     {
                         "uuid": explicit_binary,
-                        "conversation_id": {
-                            "$ref": "conversations",
-                            "$id": ObjectId(conversation_id),
-                        },
+                        "conversation_id": ObjectId(conversation_id),
                         "isPromptable": is_promptable,
                         "first_n_queries": current_n_queries,
                         "user": user,
@@ -233,10 +219,12 @@ class GetPubliclySharedConversations(Resource):
             if (
                 shared
                 and "conversation_id" in shared
-                and isinstance(shared["conversation_id"], DBRef)
             ):
-                conversation_ref = shared["conversation_id"]
-                conversation = db.dereference(conversation_ref)
+                # conversation_id is now stored as an ObjectId, not a DBRef
+                conversation_id = shared["conversation_id"]
+                conversation = conversations_collection.find_one(
+                    {"_id": conversation_id}
+                )
                 if conversation is None:
                     return make_response(
                         jsonify(

application/api/user/tools/routes.py

@@ -56,9 +56,10 @@ class GetTools(Resource):
             tools = user_tools_collection.find({"user": user})
             user_tools = []
             for tool in tools:
-                tool["id"] = str(tool["_id"])
-                tool.pop("_id")
-                user_tools.append(tool)
+                tool_copy = {**tool}
+                tool_copy["id"] = str(tool["_id"])
+                tool_copy.pop("_id", None)
+                user_tools.append(tool_copy)
         except Exception as err:
             current_app.logger.error(f"Error getting user tools: {err}", exc_info=True)
             return make_response(jsonify({"success": False}), 400)

application/core/model_configs.py

@@ -0,0 +1,189 @@
+"""
+Model configurations for all supported LLM providers.
+"""
+
+from application.core.model_settings import (
+    AvailableModel,
+    ModelCapabilities,
+    ModelProvider,
+)
+
+OPENAI_ATTACHMENTS = [
+    "application/pdf",
+    "image/png",
+    "image/jpeg",
+    "image/jpg",
+    "image/webp",
+    "image/gif",
+]
+
+GOOGLE_ATTACHMENTS = [
+    "application/pdf",
+    "image/png",
+    "image/jpeg",
+    "image/jpg",
+    "image/webp",
+    "image/gif",
+]
+
+
+OPENAI_MODELS = [
+    AvailableModel(
+        id="gpt-5.1",
+        provider=ModelProvider.OPENAI,
+        display_name="GPT-5.1",
+        description="Flagship model with enhanced reasoning, coding, and agentic capabilities",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=OPENAI_ATTACHMENTS,
+            context_window=200000,
+        ),
+    ),
+    AvailableModel(
+        id="gpt-5-mini",
+        provider=ModelProvider.OPENAI,
+        display_name="GPT-5 Mini",
+        description="Faster, cost-effective variant of GPT-5.1",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=OPENAI_ATTACHMENTS,
+            context_window=200000,
+        ),
+    )
+]
+
+
+ANTHROPIC_MODELS = [
+    AvailableModel(
+        id="claude-3-5-sonnet-20241022",
+        provider=ModelProvider.ANTHROPIC,
+        display_name="Claude 3.5 Sonnet (Latest)",
+        description="Latest Claude 3.5 Sonnet with enhanced capabilities",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=200000,
+        ),
+    ),
+    AvailableModel(
+        id="claude-3-5-sonnet",
+        provider=ModelProvider.ANTHROPIC,
+        display_name="Claude 3.5 Sonnet",
+        description="Balanced performance and capability",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=200000,
+        ),
+    ),
+    AvailableModel(
+        id="claude-3-opus",
+        provider=ModelProvider.ANTHROPIC,
+        display_name="Claude 3 Opus",
+        description="Most capable Claude model",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=200000,
+        ),
+    ),
+    AvailableModel(
+        id="claude-3-haiku",
+        provider=ModelProvider.ANTHROPIC,
+        display_name="Claude 3 Haiku",
+        description="Fastest Claude model",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=200000,
+        ),
+    ),
+]
+
+
+GOOGLE_MODELS = [
+    AvailableModel(
+        id="gemini-flash-latest",
+        provider=ModelProvider.GOOGLE,
+        display_name="Gemini Flash (Latest)",
+        description="Latest experimental Gemini model",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=GOOGLE_ATTACHMENTS,
+            context_window=int(1e6),
+        ),
+    ),
+    AvailableModel(
+        id="gemini-flash-lite-latest",
+        provider=ModelProvider.GOOGLE,
+        display_name="Gemini Flash Lite (Latest)",
+        description="Fast with huge context window",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=GOOGLE_ATTACHMENTS,
+            context_window=int(1e6),
+        ),
+    ),
+    AvailableModel(
+        id="gemini-3-pro-preview",
+        provider=ModelProvider.GOOGLE,
+        display_name="Gemini 3 Pro",
+        description="Most capable Gemini model",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=GOOGLE_ATTACHMENTS,
+            context_window=2000000,
+        ),
+    ),
+]
+
+
+GROQ_MODELS = [
+    AvailableModel(
+        id="llama-3.3-70b-versatile",
+        provider=ModelProvider.GROQ,
+        display_name="Llama 3.3 70B",
+        description="Latest Llama model with high-speed inference",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=128000,
+        ),
+    ),
+    AvailableModel(
+        id="llama-3.1-8b-instant",
+        provider=ModelProvider.GROQ,
+        display_name="Llama 3.1 8B",
+        description="Ultra-fast inference",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=128000,
+        ),
+    ),
+    AvailableModel(
+        id="mixtral-8x7b-32768",
+        provider=ModelProvider.GROQ,
+        display_name="Mixtral 8x7B",
+        description="High-speed inference with tools",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            context_window=32768,
+        ),
+    ),
+]
+
+
+AZURE_OPENAI_MODELS = [
+    AvailableModel(
+        id="azure-gpt-4",
+        provider=ModelProvider.AZURE_OPENAI,
+        display_name="Azure OpenAI GPT-4",
+        description="Azure-hosted GPT model",
+        capabilities=ModelCapabilities(
+            supports_tools=True,
+            supports_structured_output=True,
+            supported_attachment_types=OPENAI_ATTACHMENTS,
+            context_window=8192,
+        ),
+    ),
+]

application/core/model_settings.py

@@ -0,0 +1,236 @@
+import logging
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+class ModelProvider(str, Enum):
+    OPENAI = "openai"
+    AZURE_OPENAI = "azure_openai"
+    ANTHROPIC = "anthropic"
+    GROQ = "groq"
+    GOOGLE = "google"
+    HUGGINGFACE = "huggingface"
+    LLAMA_CPP = "llama.cpp"
+    DOCSGPT = "docsgpt"
+    PREMAI = "premai"
+    SAGEMAKER = "sagemaker"
+    NOVITA = "novita"
+
+
+@dataclass
+class ModelCapabilities:
+    supports_tools: bool = False
+    supports_structured_output: bool = False
+    supports_streaming: bool = True
+    supported_attachment_types: List[str] = field(default_factory=list)
+    context_window: int = 128000
+    input_cost_per_token: Optional[float] = None
+    output_cost_per_token: Optional[float] = None
+
+
+@dataclass
+class AvailableModel:
+    id: str
+    provider: ModelProvider
+    display_name: str
+    description: str = ""
+    capabilities: ModelCapabilities = field(default_factory=ModelCapabilities)
+    enabled: bool = True
+    base_url: Optional[str] = None
+
+    def to_dict(self) -> Dict:
+        result = {
+            "id": self.id,
+            "provider": self.provider.value,
+            "display_name": self.display_name,
+            "description": self.description,
+            "supported_attachment_types": self.capabilities.supported_attachment_types,
+            "supports_tools": self.capabilities.supports_tools,
+            "supports_structured_output": self.capabilities.supports_structured_output,
+            "supports_streaming": self.capabilities.supports_streaming,
+            "context_window": self.capabilities.context_window,
+            "enabled": self.enabled,
+        }
+        if self.base_url:
+            result["base_url"] = self.base_url
+        return result
+
+
+class ModelRegistry:
+    _instance = None
+    _initialized = False
+
+    def __new__(cls):
+        if cls._instance is None:
+            cls._instance = super().__new__(cls)
+        return cls._instance
+
+    def __init__(self):
+        if not ModelRegistry._initialized:
+            self.models: Dict[str, AvailableModel] = {}
+            self.default_model_id: Optional[str] = None
+            self._load_models()
+            ModelRegistry._initialized = True
+
+    @classmethod
+    def get_instance(cls) -> "ModelRegistry":
+        return cls()
+
+    def _load_models(self):
+        from application.core.settings import settings
+
+        self.models.clear()
+
+        self._add_docsgpt_models(settings)
+        if settings.OPENAI_API_KEY or (
+            settings.LLM_PROVIDER == "openai" and settings.API_KEY
+        ):
+            self._add_openai_models(settings)
+        if settings.OPENAI_API_BASE or (
+            settings.LLM_PROVIDER == "azure_openai" and settings.API_KEY
+        ):
+            self._add_azure_openai_models(settings)
+        if settings.ANTHROPIC_API_KEY or (
+            settings.LLM_PROVIDER == "anthropic" and settings.API_KEY
+        ):
+            self._add_anthropic_models(settings)
+        if settings.GOOGLE_API_KEY or (
+            settings.LLM_PROVIDER == "google" and settings.API_KEY
+        ):
+            self._add_google_models(settings)
+        if settings.GROQ_API_KEY or (
+            settings.LLM_PROVIDER == "groq" and settings.API_KEY
+        ):
+            self._add_groq_models(settings)
+        if settings.HUGGINGFACE_API_KEY or (
+            settings.LLM_PROVIDER == "huggingface" and settings.API_KEY
+        ):
+            self._add_huggingface_models(settings)
+        # Default model selection
+
+        if settings.LLM_NAME and settings.LLM_NAME in self.models:
+            self.default_model_id = settings.LLM_NAME
+        elif settings.LLM_PROVIDER and settings.API_KEY:
+            for model_id, model in self.models.items():
+                if model.provider.value == settings.LLM_PROVIDER:
+                    self.default_model_id = model_id
+                    break
+        else:
+            self.default_model_id = next(iter(self.models.keys()))
+        logger.info(
+            f"ModelRegistry loaded {len(self.models)} models, default: {self.default_model_id}"
+        )
+
+    def _add_openai_models(self, settings):
+        from application.core.model_configs import OPENAI_MODELS
+
+        if settings.OPENAI_API_KEY:
+            for model in OPENAI_MODELS:
+                self.models[model.id] = model
+            return
+        if settings.LLM_PROVIDER == "openai" and settings.LLM_NAME:
+            for model in OPENAI_MODELS:
+                if model.id == settings.LLM_NAME:
+                    self.models[model.id] = model
+                    return
+        for model in OPENAI_MODELS:
+            self.models[model.id] = model
+
+    def _add_azure_openai_models(self, settings):
+        from application.core.model_configs import AZURE_OPENAI_MODELS
+
+        if settings.LLM_PROVIDER == "azure_openai" and settings.LLM_NAME:
+            for model in AZURE_OPENAI_MODELS:
+                if model.id == settings.LLM_NAME:
+                    self.models[model.id] = model
+                    return
+        for model in AZURE_OPENAI_MODELS:
+            self.models[model.id] = model
+
+    def _add_anthropic_models(self, settings):
+        from application.core.model_configs import ANTHROPIC_MODELS
+
+        if settings.ANTHROPIC_API_KEY:
+            for model in ANTHROPIC_MODELS:
+                self.models[model.id] = model
+            return
+        if settings.LLM_PROVIDER == "anthropic" and settings.LLM_NAME:
+            for model in ANTHROPIC_MODELS:
+                if model.id == settings.LLM_NAME:
+                    self.models[model.id] = model
+                    return
+        for model in ANTHROPIC_MODELS:
+            self.models[model.id] = model
+
+    def _add_google_models(self, settings):
+        from application.core.model_configs import GOOGLE_MODELS
+
+        if settings.GOOGLE_API_KEY:
+            for model in GOOGLE_MODELS:
+                self.models[model.id] = model
+            return
+        if settings.LLM_PROVIDER == "google" and settings.LLM_NAME:
+            for model in GOOGLE_MODELS:
+                if model.id == settings.LLM_NAME:
+                    self.models[model.id] = model
+                    return
+        for model in GOOGLE_MODELS:
+            self.models[model.id] = model
+
+    def _add_groq_models(self, settings):
+        from application.core.model_configs import GROQ_MODELS
+
+        if settings.GROQ_API_KEY:
+            for model in GROQ_MODELS:
+                self.models[model.id] = model
+            return
+        if settings.LLM_PROVIDER == "groq" and settings.LLM_NAME:
+            for model in GROQ_MODELS:
+                if model.id == settings.LLM_NAME:
+                    self.models[model.id] = model
+                    return
+        for model in GROQ_MODELS:
+            self.models[model.id] = model
+
+    def _add_docsgpt_models(self, settings):
+        model_id = "docsgpt-local"
+        model = AvailableModel(
+            id=model_id,
+            provider=ModelProvider.DOCSGPT,
+            display_name="DocsGPT Model",
+            description="Local model",
+            capabilities=ModelCapabilities(
+                supports_tools=False,
+                supported_attachment_types=[],
+            ),
+        )
+        self.models[model_id] = model
+
+    def _add_huggingface_models(self, settings):
+        model_id = "huggingface-local"
+        model = AvailableModel(
+            id=model_id,
+            provider=ModelProvider.HUGGINGFACE,
+            display_name="Hugging Face Model",
+            description="Local Hugging Face model",
+            capabilities=ModelCapabilities(
+                supports_tools=False,
+                supported_attachment_types=[],
+            ),
+        )
+        self.models[model_id] = model
+
+    def get_model(self, model_id: str) -> Optional[AvailableModel]:
+        return self.models.get(model_id)
+
+    def get_all_models(self) -> List[AvailableModel]:
+        return list(self.models.values())
+
+    def get_enabled_models(self) -> List[AvailableModel]:
+        return [m for m in self.models.values() if m.enabled]
+
+    def model_exists(self, model_id: str) -> bool:
+        return model_id in self.models

application/core/model_utils.py

@@ -0,0 +1,91 @@
+from typing import Any, Dict, Optional
+
+from application.core.model_settings import ModelRegistry
+
+
+def get_api_key_for_provider(provider: str) -> Optional[str]:
+    """Get the appropriate API key for a provider"""
+    from application.core.settings import settings
+
+    provider_key_map = {
+        "openai": settings.OPENAI_API_KEY,
+        "anthropic": settings.ANTHROPIC_API_KEY,
+        "google": settings.GOOGLE_API_KEY,
+        "groq": settings.GROQ_API_KEY,
+        "huggingface": settings.HUGGINGFACE_API_KEY,
+        "azure_openai": settings.API_KEY,
+        "docsgpt": None,
+        "llama.cpp": None,
+    }
+
+    provider_key = provider_key_map.get(provider)
+    if provider_key:
+        return provider_key
+    return settings.API_KEY
+
+
+def get_all_available_models() -> Dict[str, Dict[str, Any]]:
+    """Get all available models with metadata for API response"""
+    registry = ModelRegistry.get_instance()
+    return {model.id: model.to_dict() for model in registry.get_enabled_models()}
+
+
+def validate_model_id(model_id: str) -> bool:
+    """Check if a model ID exists in registry"""
+    registry = ModelRegistry.get_instance()
+    return registry.model_exists(model_id)
+
+
+def get_model_capabilities(model_id: str) -> Optional[Dict[str, Any]]:
+    """Get capabilities for a specific model"""
+    registry = ModelRegistry.get_instance()
+    model = registry.get_model(model_id)
+    if model:
+        return {
+            "supported_attachment_types": model.capabilities.supported_attachment_types,
+            "supports_tools": model.capabilities.supports_tools,
+            "supports_structured_output": model.capabilities.supports_structured_output,
+            "context_window": model.capabilities.context_window,
+        }
+    return None
+
+
+def get_default_model_id() -> str:
+    """Get the system default model ID"""
+    registry = ModelRegistry.get_instance()
+    return registry.default_model_id
+
+
+def get_provider_from_model_id(model_id: str) -> Optional[str]:
+    """Get the provider name for a given model_id"""
+    registry = ModelRegistry.get_instance()
+    model = registry.get_model(model_id)
+    if model:
+        return model.provider.value
+    return None
+
+
+def get_token_limit(model_id: str) -> int:
+    """
+    Get context window (token limit) for a model.
+    Returns model's context_window or default 128000 if model not found.
+    """
+    from application.core.settings import settings
+
+    registry = ModelRegistry.get_instance()
+    model = registry.get_model(model_id)
+    if model:
+        return model.capabilities.context_window
+    return settings.DEFAULT_LLM_TOKEN_LIMIT
+
+
+def get_base_url_for_model(model_id: str) -> Optional[str]:
+    """
+    Get the custom base_url for a specific model if configured.
+    Returns None if no custom base_url is set.
+    """
+    registry = ModelRegistry.get_instance()
+    model = registry.get_model(model_id)
+    if model:
+        return model.base_url
+    return None

application/core/settings.py

@@ -22,11 +22,11 @@ class Settings(BaseSettings):
     MONGO_DB_NAME: str = "docsgpt"
     LLM_PATH: str = os.path.join(current_dir, "models/docsgpt-7b-f16.gguf")
     DEFAULT_MAX_HISTORY: int = 150
-    LLM_TOKEN_LIMITS: dict = {
-        "gpt-4o-mini": 128000,
-        "gpt-3.5-turbo": 4096,
-        "claude-2": 1e5,
-        "gemini-2.5-flash": 1e6,
+    DEFAULT_LLM_TOKEN_LIMIT: int = 128000  # Fallback when model not found in registry
+    RESERVED_TOKENS: dict = {
+        "system_prompt": 500,
+        "current_query": 500,
+        "safety_buffer": 1000,
     }
     DEFAULT_AGENT_LIMITS: dict = {
         "token_limit": 50000,
@@ -56,14 +56,22 @@ class Settings(BaseSettings):
     )
 
     # GitHub source
-    GITHUB_ACCESS_TOKEN: Optional[str] = None # PAT token with read repo access
+    GITHUB_ACCESS_TOKEN: Optional[str] = None  # PAT token with read repo access
 
     # LLM Cache
     CACHE_REDIS_URL: str = "redis://localhost:6379/2"
 
     API_URL: str = "http://localhost:7091"  # backend url for celery worker
 
-    API_KEY: Optional[str] = None  # LLM api key
+    API_KEY: Optional[str] = None  # LLM api key (used by LLM_PROVIDER)
+
+    # Provider-specific API keys (for multi-model support)
+    OPENAI_API_KEY: Optional[str] = None
+    ANTHROPIC_API_KEY: Optional[str] = None
+    GOOGLE_API_KEY: Optional[str] = None
+    GROQ_API_KEY: Optional[str] = None
+    HUGGINGFACE_API_KEY: Optional[str] = None
+
     EMBEDDINGS_KEY: Optional[str] = (
         None  # api key for embeddings (if using openai, just copy API_KEY)
     )
@@ -130,8 +138,19 @@ class Settings(BaseSettings):
     # Encryption settings
     ENCRYPTION_SECRET_KEY: str = "default-docsgpt-encryption-key"
 
-    TTS_PROVIDER: str = "google_tts" # google_tts or elevenlabs
+    TTS_PROVIDER: str = "google_tts"  # google_tts or elevenlabs
     ELEVENLABS_API_KEY: Optional[str] = None
 
+    # Tool pre-fetch settings
+    ENABLE_TOOL_PREFETCH: bool = True
+
+    # Conversation Compression Settings
+    ENABLE_CONVERSATION_COMPRESSION: bool = True
+    COMPRESSION_THRESHOLD_PERCENTAGE: float = 0.8  # Trigger at 80% of context
+    COMPRESSION_MODEL_OVERRIDE: Optional[str] = None  # Use different model for compression
+    COMPRESSION_PROMPT_VERSION: str = "v1.0"  # Track prompt iterations
+    COMPRESSION_MAX_HISTORY_POINTS: int = 3  # Keep only last N compression points to prevent DB bloat
+
+
 path = Path(__file__).parent.parent.absolute()
 settings = Settings(_env_file=path.joinpath(".env"), _env_file_encoding="utf-8")

application/llm/anthropic.py

@@ -1,30 +1,41 @@
-from application.llm.base import BaseLLM
+from anthropic import AI_PROMPT, Anthropic, HUMAN_PROMPT
+
 from application.core.settings import settings
+from application.llm.base import BaseLLM
 
 
 class AnthropicLLM(BaseLLM):
 
-    def __init__(self, api_key=None, user_api_key=None, *args, **kwargs):
-        from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
+    def __init__(self, api_key=None, user_api_key=None, base_url=None, *args, **kwargs):
 
         super().__init__(*args, **kwargs)
-        self.api_key = (
-            api_key or settings.ANTHROPIC_API_KEY
-        )  # If not provided, use a default from settings
+        self.api_key = api_key or settings.ANTHROPIC_API_KEY or settings.API_KEY
         self.user_api_key = user_api_key
-        self.anthropic = Anthropic(api_key=self.api_key)
+
+        # Use custom base_url if provided
+        if base_url:
+            self.anthropic = Anthropic(api_key=self.api_key, base_url=base_url)
+        else:
+            self.anthropic = Anthropic(api_key=self.api_key)
+
         self.HUMAN_PROMPT = HUMAN_PROMPT
         self.AI_PROMPT = AI_PROMPT
 
     def _raw_gen(
-        self, baseself, model, messages, stream=False, tools=None, max_tokens=300, **kwargs
+        self,
+        baseself,
+        model,
+        messages,
+        stream=False,
+        tools=None,
+        max_tokens=300,
+        **kwargs,
     ):
         context = messages[0]["content"]
         user_question = messages[-1]["content"]
         prompt = f"### Context \n {context} \n ### Question \n {user_question}"
         if stream:
             return self.gen_stream(model, prompt, stream, max_tokens, **kwargs)
-
         completion = self.anthropic.completions.create(
             model=model,
             max_tokens_to_sample=max_tokens,
@@ -34,7 +45,14 @@ class AnthropicLLM(BaseLLM):
         return completion.completion
 
     def _raw_gen_stream(
-        self, baseself, model, messages, stream=True, tools=None, max_tokens=300, **kwargs
+        self,
+        baseself,
+        model,
+        messages,
+        stream=True,
+        tools=None,
+        max_tokens=300,
+        **kwargs,
     ):
         context = messages[0]["content"]
         user_question = messages[-1]["content"]
@@ -50,5 +68,5 @@ class AnthropicLLM(BaseLLM):
             for completion in stream_response:
                 yield completion.completion
         finally:
-            if hasattr(stream_response, 'close'):
+            if hasattr(stream_response, "close"):
                 stream_response.close()

application/llm/base.py

@@ -13,30 +13,32 @@ class BaseLLM(ABC):
     def __init__(
         self,
         decoded_token=None,
+        model_id=None,
+        base_url=None,
     ):
         self.decoded_token = decoded_token
+        self.model_id = model_id
+        self.base_url = base_url
         self.token_usage = {"prompt_tokens": 0, "generated_tokens": 0}
-        self.fallback_provider = settings.FALLBACK_LLM_PROVIDER
-        self.fallback_model_name = settings.FALLBACK_LLM_NAME
-        self.fallback_llm_api_key = settings.FALLBACK_LLM_API_KEY
         self._fallback_llm = None
+        self._fallback_sequence_index = 0
 
     @property
     def fallback_llm(self):
-        """Lazy-loaded fallback LLM instance."""
-        if (
-            self._fallback_llm is None
-            and self.fallback_provider
-            and self.fallback_model_name
-        ):
+        """Lazy-loaded fallback LLM from FALLBACK_* settings."""
+        if self._fallback_llm is None and settings.FALLBACK_LLM_PROVIDER:
             try:
                 from application.llm.llm_creator import LLMCreator
 
                 self._fallback_llm = LLMCreator.create_llm(
-                    self.fallback_provider,
-                    self.fallback_llm_api_key,
-                    None,
-                    self.decoded_token,
+                    settings.FALLBACK_LLM_PROVIDER,
+                    api_key=settings.FALLBACK_LLM_API_KEY or settings.API_KEY,
+                    user_api_key=None,
+                    decoded_token=self.decoded_token,
+                    model_id=settings.FALLBACK_LLM_NAME,
+                )
+                logger.info(
+                    f"Fallback LLM initialized: {settings.FALLBACK_LLM_PROVIDER}/{settings.FALLBACK_LLM_NAME}"
                 )
             except Exception as e:
                 logger.error(
@@ -44,11 +46,17 @@ class BaseLLM(ABC):
                 )
         return self._fallback_llm
 
+    @staticmethod
+    def _remove_null_values(args_dict):
+        if not isinstance(args_dict, dict):
+            return args_dict
+        return {k: v for k, v in args_dict.items() if v is not None}
+
     def _execute_with_fallback(
         self, method_name: str, decorators: list, *args, **kwargs
     ):
         """
-        Unified method execution with fallback support.
+        Execute method with fallback support.
 
         Args:
             method_name: Name of the raw method ('_raw_gen' or '_raw_gen_stream')
@@ -67,10 +75,10 @@ class BaseLLM(ABC):
             return decorated_method()
         except Exception as e:
             if not self.fallback_llm:
-                logger.error(f"Primary LLM failed and no fallback available: {str(e)}")
+                logger.error(f"Primary LLM failed and no fallback configured: {str(e)}")
                 raise
             logger.warning(
-                f"Falling back to {self.fallback_provider}/{self.fallback_model_name}. Error: {str(e)}"
+                f"Primary LLM failed. Falling back to {settings.FALLBACK_LLM_PROVIDER}/{settings.FALLBACK_LLM_NAME}. Error: {str(e)}"
             )
 
             fallback_method = getattr(

application/llm/docsgpt_provider.py

@@ -1,5 +1,7 @@
 import json
 
+from openai import OpenAI
+
 from application.core.settings import settings
 from application.llm.base import BaseLLM
 
@@ -7,12 +9,11 @@ from application.llm.base import BaseLLM
 class DocsGPTAPILLM(BaseLLM):
 
     def __init__(self, api_key=None, user_api_key=None, *args, **kwargs):
-        from openai import OpenAI
 
         super().__init__(*args, **kwargs)
-        self.client = OpenAI(api_key="sk-docsgpt-public", base_url="https://oai.arc53.com")
+        self.api_key = "sk-docsgpt-public"
+        self.client = OpenAI(api_key=self.api_key, base_url="https://oai.arc53.com")
         self.user_api_key = user_api_key
-        self.api_key = api_key
 
     def _clean_messages_openai(self, messages):
         cleaned_messages = []
@@ -22,7 +23,6 @@ class DocsGPTAPILLM(BaseLLM):
 
             if role == "model":
                 role = "assistant"
-
             if role and content is not None:
                 if isinstance(content, str):
                     cleaned_messages.append({"role": role, "content": content})
@@ -33,14 +33,15 @@ class DocsGPTAPILLM(BaseLLM):
                                 {"role": role, "content": item["text"]}
                             )
                         elif "function_call" in item:
+                            cleaned_args = self._remove_null_values(
+                                item["function_call"]["args"]
+                            )
                             tool_call = {
                                 "id": item["function_call"]["call_id"],
                                 "type": "function",
                                 "function": {
                                     "name": item["function_call"]["name"],
-                                    "arguments": json.dumps(
-                                        item["function_call"]["args"]
-                                    ),
+                                    "arguments": json.dumps(cleaned_args),
                                 },
                             }
                             cleaned_messages.append(
@@ -68,7 +69,6 @@ class DocsGPTAPILLM(BaseLLM):
                             )
                 else:
                     raise ValueError(f"Unexpected content type: {type(content)}")
-
         return cleaned_messages
 
     def _raw_gen(
@@ -120,7 +120,6 @@ class DocsGPTAPILLM(BaseLLM):
             response = self.client.chat.completions.create(
                 model="docsgpt", messages=messages, stream=stream, **kwargs
             )
-
         try:
             for line in response:
                 if (
@@ -132,8 +131,8 @@ class DocsGPTAPILLM(BaseLLM):
                 elif len(line.choices) > 0:
                     yield line.choices[0]
         finally:
-            if hasattr(response, 'close'):
+            if hasattr(response, "close"):
                 response.close()
 
     def _supports_tools(self):
-        return True
+        return True

application/llm/google_ai.py

@@ -1,4 +1,3 @@
-import json
 import logging
 
 from google import genai
@@ -11,10 +10,13 @@ from application.storage.storage_creator import StorageCreator
 
 
 class GoogleLLM(BaseLLM):
-    def __init__(self, api_key=None, user_api_key=None, *args, **kwargs):
+    def __init__(
+        self, api_key=None, user_api_key=None, decoded_token=None, *args, **kwargs
+    ):
         super().__init__(*args, **kwargs)
-        self.api_key = api_key
+        self.api_key = api_key or settings.GOOGLE_API_KEY or settings.API_KEY
         self.user_api_key = user_api_key
+
         self.client = genai.Client(api_key=self.api_key)
         self.storage = StorageCreator.get_storage()
 
@@ -32,6 +34,12 @@ class GoogleLLM(BaseLLM):
             "image/jpg",
             "image/webp",
             "image/gif",
+            "application/pdf",
+            "image/png",
+            "image/jpeg",
+            "image/jpg",
+            "image/webp",
+            "image/gif",
         ]
 
     def prepare_messages_with_attachments(self, messages, attachments=None):
@@ -47,21 +55,19 @@ class GoogleLLM(BaseLLM):
         """
         if not attachments:
             return messages
-
         prepared_messages = messages.copy()
 
         # Find the user message to attach files to the last one
+
         user_message_index = None
         for i in range(len(prepared_messages) - 1, -1, -1):
             if prepared_messages[i].get("role") == "user":
                 user_message_index = i
                 break
-
         if user_message_index is None:
             user_message = {"role": "user", "content": []}
             prepared_messages.append(user_message)
             user_message_index = len(prepared_messages) - 1
-
         if isinstance(prepared_messages[user_message_index].get("content"), str):
             text_content = prepared_messages[user_message_index]["content"]
             prepared_messages[user_message_index]["content"] = [
@@ -69,7 +75,6 @@ class GoogleLLM(BaseLLM):
             ]
         elif not isinstance(prepared_messages[user_message_index].get("content"), list):
             prepared_messages[user_message_index]["content"] = []
-
         files = []
         for attachment in attachments:
             mime_type = attachment.get("mime_type")
@@ -92,11 +97,9 @@ class GoogleLLM(BaseLLM):
                                 "text": f"[File could not be processed: {attachment.get('path', 'unknown')}]",
                             }
                         )
-
         if files:
             logging.info(f"GoogleLLM: Adding {len(files)} files to message")
             prepared_messages[user_message_index]["content"].append({"files": files})
-
         return prepared_messages
 
     def _upload_file_to_google(self, attachment):
@@ -111,14 +114,11 @@ class GoogleLLM(BaseLLM):
         """
         if "google_file_uri" in attachment:
             return attachment["google_file_uri"]
-
         file_path = attachment.get("path")
         if not file_path:
             raise ValueError("No file path provided in attachment")
-
         if not self.storage.file_exists(file_path):
             raise FileNotFoundError(f"File not found: {file_path}")
-
         try:
             file_uri = self.storage.process_file(
                 file_path,
@@ -136,24 +136,48 @@ class GoogleLLM(BaseLLM):
                 attachments_collection.update_one(
                     {"_id": attachment["_id"]}, {"$set": {"google_file_uri": file_uri}}
                 )
-
             return file_uri
         except Exception as e:
             logging.error(f"Error uploading file to Google AI: {e}", exc_info=True)
             raise
 
     def _clean_messages_google(self, messages):
-        """Convert OpenAI format messages to Google AI format."""
+        """
+        Convert OpenAI format messages to Google AI format and collect system prompts.
+
+        Returns:
+            tuple[list[types.Content], Optional[str]]: cleaned messages and optional
+            combined system instruction.
+        """
         cleaned_messages = []
+        system_instructions = []
+
+        def _extract_system_text(content):
+            if isinstance(content, str):
+                return content
+            if isinstance(content, list):
+                parts = []
+                for item in content:
+                    if isinstance(item, dict) and "text" in item and item["text"] is not None:
+                        parts.append(item["text"])
+                return "\n".join(parts)
+            return ""
+
         for message in messages:
             role = message.get("role")
             content = message.get("content")
 
+            # Gemini only accepts user/model in the contents list.
+            if role == "system":
+                sys_text = _extract_system_text(content)
+                if sys_text:
+                    system_instructions.append(sys_text)
+                continue
+
             if role == "assistant":
                 role = "model"
             elif role == "tool":
                 role = "model"
-
             parts = []
             if role and content is not None:
                 if isinstance(content, str):
@@ -163,12 +187,32 @@ class GoogleLLM(BaseLLM):
                         if "text" in item:
                             parts.append(types.Part.from_text(text=item["text"]))
                         elif "function_call" in item:
-                            parts.append(
-                                types.Part.from_function_call(
-                                    name=item["function_call"]["name"],
-                                    args=item["function_call"]["args"],
-                                )
+                            # Remove null values from args to avoid API errors
+
+                            cleaned_args = self._remove_null_values(
+                                item["function_call"]["args"]
                             )
+                            # Create function call part with thought_signature if present
+                            # For Gemini 3 models, we need to include thought_signature
+                            if "thought_signature" in item:
+                                # Use Part constructor with functionCall and thoughtSignature
+                                parts.append(
+                                    types.Part(
+                                        functionCall=types.FunctionCall(
+                                            name=item["function_call"]["name"],
+                                            args=cleaned_args,
+                                        ),
+                                        thoughtSignature=item["thought_signature"],
+                                    )
+                                )
+                            else:
+                                # Use helper method when no thought_signature
+                                parts.append(
+                                    types.Part.from_function_call(
+                                        name=item["function_call"]["name"],
+                                        args=cleaned_args,
+                                    )
+                                )
                         elif "function_response" in item:
                             parts.append(
                                 types.Part.from_function_response(
@@ -190,11 +234,10 @@ class GoogleLLM(BaseLLM):
                             )
                 else:
                     raise ValueError(f"Unexpected content type: {type(content)}")
-
                 if parts:
                     cleaned_messages.append(types.Content(role=role, parts=parts))
-
-        return cleaned_messages
+        system_instruction = "\n\n".join(system_instructions) if system_instructions else None
+        return cleaned_messages, system_instruction
 
     def _clean_schema(self, schema_obj):
         """
@@ -229,8 +272,8 @@ class GoogleLLM(BaseLLM):
                 cleaned[key] = [self._clean_schema(item) for item in value]
             else:
                 cleaned[key] = value
-
         # Validate that required properties actually exist in properties
+
         if "required" in cleaned and "properties" in cleaned:
             valid_required = []
             properties_keys = set(cleaned["properties"].keys())
@@ -243,7 +286,6 @@ class GoogleLLM(BaseLLM):
                 cleaned.pop("required", None)
         elif "required" in cleaned and "properties" not in cleaned:
             cleaned.pop("required", None)
-
         return cleaned
 
     def _clean_tools_format(self, tools_list):
@@ -259,7 +301,6 @@ class GoogleLLM(BaseLLM):
                     cleaned_properties = {}
                     for k, v in properties.items():
                         cleaned_properties[k] = self._clean_schema(v)
-
                     genai_function = dict(
                         name=function["name"],
                         description=function["description"],
@@ -278,12 +319,65 @@ class GoogleLLM(BaseLLM):
                         name=function["name"],
                         description=function["description"],
                     )
-
                 genai_tool = types.Tool(function_declarations=[genai_function])
                 genai_tools.append(genai_tool)
-
         return genai_tools
 
+    def _extract_preview_from_message(self, message):
+        """Get a short, human-readable preview from the last message."""
+        try:
+            if hasattr(message, "parts"):
+                for part in reversed(message.parts):
+                    if getattr(part, "text", None):
+                        return part.text
+                    function_call = getattr(part, "function_call", None)
+                    if function_call:
+                        name = getattr(function_call, "name", "") or "function_call"
+                        return f"function_call:{name}"
+                    function_response = getattr(part, "function_response", None)
+                    if function_response:
+                        name = getattr(function_response, "name", "") or "function_response"
+                        return f"function_response:{name}"
+            if isinstance(message, dict):
+                content = message.get("content")
+                if isinstance(content, str):
+                    return content
+                if isinstance(content, list):
+                    for item in reversed(content):
+                        if isinstance(item, str):
+                            return item
+                        if isinstance(item, dict):
+                            if item.get("text"):
+                                return item["text"]
+                            if item.get("function_call"):
+                                fn = item["function_call"]
+                                if isinstance(fn, dict):
+                                    name = fn.get("name") or "function_call"
+                                    return f"function_call:{name}"
+                                return "function_call"
+                            if item.get("function_response"):
+                                resp = item["function_response"]
+                                if isinstance(resp, dict):
+                                    name = resp.get("name") or "function_response"
+                                    return f"function_response:{name}"
+                                return "function_response"
+                if "text" in message and isinstance(message["text"], str):
+                    return message["text"]
+        except Exception:
+            pass
+        return str(message)
+
+    def _summarize_messages_for_log(self, messages, preview_chars=20):
+        """Return a compact summary for logging to avoid huge payloads."""
+        message_count = len(messages) if messages else 0
+        last_preview = ""
+        if messages:
+            last_preview = self._extract_preview_from_message(messages[-1]) or ""
+        last_preview = str(last_preview).replace("\n", " ")
+        if len(last_preview) > preview_chars:
+            last_preview = f"{last_preview[:preview_chars]}..."
+        return f"count={message_count}, last='{last_preview}'"
+
     def _raw_gen(
         self,
         baseself,
@@ -297,22 +391,20 @@ class GoogleLLM(BaseLLM):
     ):
         """Generate content using Google AI API without streaming."""
         client = genai.Client(api_key=self.api_key)
+        system_instruction = None
         if formatting == "openai":
-            messages = self._clean_messages_google(messages)
+            messages, system_instruction = self._clean_messages_google(messages)
         config = types.GenerateContentConfig()
-        if messages[0].role == "system":
-            config.system_instruction = messages[0].parts[0].text
-            messages = messages[1:]
-
+        if system_instruction:
+            config.system_instruction = system_instruction
         if tools:
             cleaned_tools = self._clean_tools_format(tools)
             config.tools = cleaned_tools
-
         # Add response schema for structured output if provided
+
         if response_schema:
             config.response_schema = response_schema
             config.response_mime_type = "application/json"
-
         response = client.models.generate_content(
             model=model,
             contents=messages,
@@ -337,23 +429,22 @@ class GoogleLLM(BaseLLM):
     ):
         """Generate content using Google AI API with streaming."""
         client = genai.Client(api_key=self.api_key)
+        system_instruction = None
         if formatting == "openai":
-            messages = self._clean_messages_google(messages)
+            messages, system_instruction = self._clean_messages_google(messages)
         config = types.GenerateContentConfig()
-        if messages[0].role == "system":
-            config.system_instruction = messages[0].parts[0].text
-            messages = messages[1:]
-
+        if system_instruction:
+            config.system_instruction = system_instruction
         if tools:
             cleaned_tools = self._clean_tools_format(tools)
             config.tools = cleaned_tools
-
         # Add response schema for structured output if provided
+
         if response_schema:
             config.response_schema = response_schema
             config.response_mime_type = "application/json"
-
         # Check if we have both tools and file attachments
+
         has_attachments = False
         for message in messages:
             for part in message.parts:
@@ -362,9 +453,12 @@ class GoogleLLM(BaseLLM):
                     break
             if has_attachments:
                 break
-
+        messages_summary = self._summarize_messages_for_log(messages)
         logging.info(
-            f"GoogleLLM: Starting stream generation. Model: {model}, Messages: {json.dumps(messages, default=str)}, Has attachments: {has_attachments}"
+            "GoogleLLM: Starting stream generation. Model: %s, Messages: %s, Has attachments: %s",
+            model,
+            messages_summary,
+            has_attachments,
         )
 
         response = client.models.generate_content_stream(
@@ -386,7 +480,7 @@ class GoogleLLM(BaseLLM):
                 elif hasattr(chunk, "text"):
                     yield chunk.text
         finally:
-            if hasattr(response, 'close'):
+            if hasattr(response, "close"):
                 response.close()
 
     def _supports_tools(self):
@@ -401,7 +495,6 @@ class GoogleLLM(BaseLLM):
         """Convert JSON schema to Google AI structured output format."""
         if not json_schema:
             return None
-
         type_map = {
             "object": "OBJECT",
             "array": "ARRAY",
@@ -414,12 +507,10 @@ class GoogleLLM(BaseLLM):
         def convert(schema):
             if not isinstance(schema, dict):
                 return schema
-
             result = {}
             schema_type = schema.get("type")
             if schema_type:
                 result["type"] = type_map.get(schema_type.lower(), schema_type.upper())
-
             for key in [
                 "description",
                 "nullable",
@@ -431,7 +522,6 @@ class GoogleLLM(BaseLLM):
             ]:
                 if key in schema:
                     result[key] = schema[key]
-
             if "format" in schema:
                 format_value = schema["format"]
                 if schema_type == "string":
@@ -441,21 +531,17 @@ class GoogleLLM(BaseLLM):
                         result["format"] = format_value
                 else:
                     result["format"] = format_value
-
             if "properties" in schema:
                 result["properties"] = {
                     k: convert(v) for k, v in schema["properties"].items()
                 }
                 if "propertyOrdering" not in result and result.get("type") == "OBJECT":
                     result["propertyOrdering"] = list(result["properties"].keys())
-
             if "items" in schema:
                 result["items"] = convert(schema["items"])
-
             for field in ["anyOf", "oneOf", "allOf"]:
                 if field in schema:
                     result[field] = [convert(s) for s in schema[field]]
-
             return result
 
         try:

application/llm/groq.py

@@ -1,13 +1,18 @@
-from application.llm.base import BaseLLM
 from openai import OpenAI
 
+from application.core.settings import settings
+from application.llm.base import BaseLLM
+
 
 class GroqLLM(BaseLLM):
     def __init__(self, api_key=None, user_api_key=None, *args, **kwargs):
+
         super().__init__(*args, **kwargs)
-        self.client = OpenAI(api_key=api_key, base_url="https://api.groq.com/openai/v1")
-        self.api_key = api_key
+        self.api_key = api_key or settings.GROQ_API_KEY or settings.API_KEY
         self.user_api_key = user_api_key
+        self.client = OpenAI(
+            api_key=self.api_key, base_url="https://api.groq.com/openai/v1"
+        )
 
     def _raw_gen(self, baseself, model, messages, stream=False, tools=None, **kwargs):
         if tools:

application/llm/handlers/base.py

@@ -1,4 +1,5 @@
 import logging
+import uuid
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
 from typing import Any, Dict, Generator, List, Optional, Union
@@ -16,6 +17,7 @@ class ToolCall:
     name: str
     arguments: Union[str, Dict]
     index: Optional[int] = None
+    thought_signature: Optional[str] = None
 
     @classmethod
     def from_dict(cls, data: Dict) -> "ToolCall":
@@ -178,6 +180,406 @@ class LLMHandler(ABC):
             system_msg["content"] += f"\n\n{combined_text}"
         return prepared_messages
 
+    def _prune_messages_minimal(self, messages: List[Dict]) -> Optional[List[Dict]]:
+        """
+        Build a minimal context: system prompt + latest user message only.
+        Drops all tool/function messages to shrink context aggressively.
+        """
+        system_message = next((m for m in messages if m.get("role") == "system"), None)
+        if not system_message:
+            logger.warning("Cannot prune messages minimally: missing system message.")
+            return None
+        last_non_system = None
+        for m in reversed(messages):
+            if m.get("role") == "user":
+                last_non_system = m
+                break
+            if not last_non_system and m.get("role") not in ("system", None):
+                last_non_system = m
+        if not last_non_system:
+            logger.warning("Cannot prune messages minimally: missing user/assistant messages.")
+            return None
+        logger.info("Pruning context to system + latest user/assistant message to proceed.")
+        return [system_message, last_non_system]
+
+    def _extract_text_from_content(self, content: Any) -> str:
+        """
+        Convert message content (str or list of parts) to plain text for compression.
+        """
+        if isinstance(content, str):
+            return content
+        if isinstance(content, list):
+            parts_text = []
+            for item in content:
+                if isinstance(item, dict):
+                    if "text" in item and item["text"] is not None:
+                        parts_text.append(str(item["text"]))
+                    elif "function_call" in item or "function_response" in item:
+                        # Keep serialized function calls/responses so the compressor sees actions
+                        parts_text.append(str(item))
+                    elif "files" in item:
+                        parts_text.append(str(item))
+            return "\n".join(parts_text)
+        return ""
+
+    def _build_conversation_from_messages(self, messages: List[Dict]) -> Optional[Dict]:
+        """
+        Build a conversation-like dict from current messages so we can compress
+        even when the conversation isn't persisted yet. Includes tool calls/results.
+        """
+        queries = []
+        current_prompt = None
+        current_tool_calls = {}
+
+        def _commit_query(response_text: str):
+            nonlocal current_prompt, current_tool_calls
+            if current_prompt is None and not response_text:
+                return
+            tool_calls_list = list(current_tool_calls.values())
+            queries.append(
+                {
+                    "prompt": current_prompt or "",
+                    "response": response_text,
+                    "tool_calls": tool_calls_list,
+                }
+            )
+            current_prompt = None
+            current_tool_calls = {}
+
+        for message in messages:
+            role = message.get("role")
+            content = message.get("content")
+
+            if role == "user":
+                current_prompt = self._extract_text_from_content(content)
+
+            elif role in {"assistant", "model"}:
+                # If this assistant turn contains tool calls, collect them; otherwise commit a response.
+                if isinstance(content, list):
+                    for item in content:
+                        if "function_call" in item:
+                            fc = item["function_call"]
+                            call_id = fc.get("call_id") or str(uuid.uuid4())
+                            current_tool_calls[call_id] = {
+                                "tool_name": "unknown_tool",
+                                "action_name": fc.get("name"),
+                                "arguments": fc.get("args"),
+                                "result": None,
+                                "status": "called",
+                                "call_id": call_id,
+                            }
+                        elif "function_response" in item:
+                            fr = item["function_response"]
+                            call_id = fr.get("call_id") or str(uuid.uuid4())
+                            current_tool_calls[call_id] = {
+                                "tool_name": "unknown_tool",
+                                "action_name": fr.get("name"),
+                                "arguments": None,
+                                "result": fr.get("response", {}).get("result"),
+                                "status": "completed",
+                                "call_id": call_id,
+                            }
+                    # No direct assistant text here; continue to next message
+                    continue
+
+                response_text = self._extract_text_from_content(content)
+                _commit_query(response_text)
+
+            elif role == "tool":
+                # Attach tool outputs to the latest pending tool call if possible
+                tool_text = self._extract_text_from_content(content)
+                # Attempt to parse function_response style
+                call_id = None
+                if isinstance(content, list):
+                    for item in content:
+                        if "function_response" in item and item["function_response"].get("call_id"):
+                            call_id = item["function_response"]["call_id"]
+                            break
+                if call_id and call_id in current_tool_calls:
+                    current_tool_calls[call_id]["result"] = tool_text
+                    current_tool_calls[call_id]["status"] = "completed"
+                elif queries:
+                    queries[-1].setdefault("tool_calls", []).append(
+                        {
+                            "tool_name": "unknown_tool",
+                            "action_name": "unknown_action",
+                            "arguments": {},
+                            "result": tool_text,
+                            "status": "completed",
+                        }
+                    )
+
+        # If there's an unfinished prompt with tool_calls but no response yet, commit it
+        if current_prompt is not None or current_tool_calls:
+            _commit_query(response_text="")
+
+        if not queries:
+            return None
+
+        return {
+            "queries": queries,
+            "compression_metadata": {
+                "is_compressed": False,
+                "compression_points": [],
+            },
+        }
+
+    def _rebuild_messages_after_compression(
+        self,
+        messages: List[Dict],
+        compressed_summary: Optional[str],
+        recent_queries: List[Dict],
+        include_current_execution: bool = False,
+        include_tool_calls: bool = False,
+    ) -> Optional[List[Dict]]:
+        """
+        Rebuild the message list after compression so tool execution can continue.
+
+        Delegates to MessageBuilder for the actual reconstruction.
+        """
+        from application.api.answer.services.compression.message_builder import (
+            MessageBuilder,
+        )
+
+        return MessageBuilder.rebuild_messages_after_compression(
+            messages=messages,
+            compressed_summary=compressed_summary,
+            recent_queries=recent_queries,
+            include_current_execution=include_current_execution,
+            include_tool_calls=include_tool_calls,
+        )
+
+    def _perform_mid_execution_compression(
+        self, agent, messages: List[Dict]
+    ) -> tuple[bool, Optional[List[Dict]]]:
+        """
+        Perform compression during tool execution and rebuild messages.
+
+        Uses the new orchestrator for simplified compression.
+
+        Args:
+            agent: The agent instance
+            messages: Current conversation messages
+
+        Returns:
+            (success: bool, rebuilt_messages: Optional[List[Dict]])
+        """
+        try:
+            from application.api.answer.services.compression import (
+                CompressionOrchestrator,
+            )
+            from application.api.answer.services.conversation_service import (
+                ConversationService,
+            )
+
+            conversation_service = ConversationService()
+            orchestrator = CompressionOrchestrator(conversation_service)
+
+            # Get conversation from database (may be None for new sessions)
+            conversation = conversation_service.get_conversation(
+                agent.conversation_id, agent.initial_user_id
+            )
+
+            if conversation:
+                # Merge current in-flight messages (including tool calls)
+                conversation_from_msgs = self._build_conversation_from_messages(messages)
+                if conversation_from_msgs:
+                    conversation = conversation_from_msgs
+            else:
+                logger.warning(
+                    "Could not load conversation for compression; attempting in-memory compression"
+                )
+                return self._perform_in_memory_compression(agent, messages)
+
+            # Use orchestrator to perform compression
+            result = orchestrator.compress_mid_execution(
+                conversation_id=agent.conversation_id,
+                user_id=agent.initial_user_id,
+                model_id=agent.model_id,
+                decoded_token=getattr(agent, "decoded_token", {}),
+                current_conversation=conversation,
+            )
+
+            if not result.success:
+                logger.warning(f"Mid-execution compression failed: {result.error}")
+                # Try minimal pruning as fallback
+                pruned = self._prune_messages_minimal(messages)
+                if pruned:
+                    agent.context_limit_reached = False
+                    agent.current_token_count = 0
+                    return True, pruned
+                return False, None
+
+            if not result.compression_performed:
+                logger.warning("Compression not performed")
+                return False, None
+
+            # Check if compression actually reduced tokens
+            if result.metadata:
+                if result.metadata.compressed_token_count >= result.metadata.original_token_count:
+                    logger.warning(
+                        "Compression did not reduce token count; falling back to minimal pruning"
+                    )
+                    pruned = self._prune_messages_minimal(messages)
+                    if pruned:
+                        agent.context_limit_reached = False
+                        agent.current_token_count = 0
+                        return True, pruned
+                    return False, None
+
+                logger.info(
+                    f"Mid-execution compression successful - ratio: {result.metadata.compression_ratio:.1f}x, "
+                    f"saved {result.metadata.original_token_count - result.metadata.compressed_token_count} tokens"
+                )
+
+            # Also store the compression summary as a visible message
+            if result.metadata:
+                conversation_service.append_compression_message(
+                    agent.conversation_id, result.metadata.to_dict()
+                )
+
+            # Update agent's compressed summary for downstream persistence
+            agent.compressed_summary = result.compressed_summary
+            agent.compression_metadata = result.metadata.to_dict() if result.metadata else None
+            agent.compression_saved = False
+
+            # Reset the context limit flag so tools can continue
+            agent.context_limit_reached = False
+            agent.current_token_count = 0
+
+            # Rebuild messages
+            rebuilt_messages = self._rebuild_messages_after_compression(
+                messages,
+                result.compressed_summary,
+                result.recent_queries,
+                include_current_execution=False,
+                include_tool_calls=False,
+            )
+
+            if rebuilt_messages is None:
+                return False, None
+
+            return True, rebuilt_messages
+
+        except Exception as e:
+            logger.error(
+                f"Error performing mid-execution compression: {str(e)}", exc_info=True
+            )
+            return False, None
+
+    def _perform_in_memory_compression(
+        self, agent, messages: List[Dict]
+    ) -> tuple[bool, Optional[List[Dict]]]:
+        """
+        Fallback compression path when the conversation is not yet persisted.
+
+        Uses CompressionService directly without DB persistence.
+        """
+        try:
+            from application.api.answer.services.compression.service import (
+                CompressionService,
+            )
+            from application.core.model_utils import (
+                get_api_key_for_provider,
+                get_provider_from_model_id,
+            )
+            from application.core.settings import settings
+            from application.llm.llm_creator import LLMCreator
+
+            conversation = self._build_conversation_from_messages(messages)
+            if not conversation:
+                logger.warning(
+                    "Cannot perform in-memory compression: no user/assistant turns found"
+                )
+                return False, None
+
+            compression_model = (
+                settings.COMPRESSION_MODEL_OVERRIDE
+                if settings.COMPRESSION_MODEL_OVERRIDE
+                else agent.model_id
+            )
+            provider = get_provider_from_model_id(compression_model)
+            api_key = get_api_key_for_provider(provider)
+            compression_llm = LLMCreator.create_llm(
+                provider,
+                api_key,
+                getattr(agent, "user_api_key", None),
+                getattr(agent, "decoded_token", None),
+                model_id=compression_model,
+            )
+
+            # Create service without DB persistence capability
+            compression_service = CompressionService(
+                llm=compression_llm,
+                model_id=compression_model,
+                conversation_service=None,  # No DB updates for in-memory
+            )
+
+            queries_count = len(conversation.get("queries", []))
+            compress_up_to = queries_count - 1
+
+            if compress_up_to < 0 or queries_count == 0:
+                logger.warning("Not enough queries to compress in-memory context")
+                return False, None
+
+            metadata = compression_service.compress_conversation(
+                conversation,
+                compress_up_to_index=compress_up_to,
+            )
+
+            # If compression doesn't reduce tokens, fall back to minimal pruning
+            if (
+                metadata.compressed_token_count
+                >= metadata.original_token_count
+            ):
+                logger.warning(
+                    "In-memory compression did not reduce token count; falling back to minimal pruning"
+                )
+                pruned = self._prune_messages_minimal(messages)
+                if pruned:
+                    agent.context_limit_reached = False
+                    agent.current_token_count = 0
+                    return True, pruned
+                return False, None
+
+            # Attach metadata to synthetic conversation
+            conversation["compression_metadata"] = {
+                "is_compressed": True,
+                "compression_points": [metadata.to_dict()],
+            }
+
+            compressed_summary, recent_queries = (
+                compression_service.get_compressed_context(conversation)
+            )
+
+            agent.compressed_summary = compressed_summary
+            agent.compression_metadata = metadata.to_dict()
+            agent.compression_saved = False
+            agent.context_limit_reached = False
+            agent.current_token_count = 0
+
+            rebuilt_messages = self._rebuild_messages_after_compression(
+                messages,
+                compressed_summary,
+                recent_queries,
+                include_current_execution=False,
+                include_tool_calls=False,
+            )
+            if rebuilt_messages is None:
+                return False, None
+
+            logger.info(
+                f"In-memory compression successful - ratio: {metadata.compression_ratio:.1f}x, "
+                f"saved {metadata.original_token_count - metadata.compressed_token_count} tokens"
+            )
+            return True, rebuilt_messages
+
+        except Exception as e:
+            logger.error(
+                f"Error performing in-memory compression: {str(e)}", exc_info=True
+            )
+            return False, None
+
     def handle_tool_calls(
         self, agent, tool_calls: List[ToolCall], tools_dict: Dict, messages: List[Dict]
     ) -> Generator:
@@ -195,7 +597,110 @@ class LLMHandler(ABC):
         """
         updated_messages = messages.copy()
 
-        for call in tool_calls:
+        for i, call in enumerate(tool_calls):
+            # Check context limit before executing tool call
+            if hasattr(agent, '_check_context_limit') and agent._check_context_limit(updated_messages):
+                # Context limit reached - attempt mid-execution compression
+                compression_attempted = False
+                compression_successful = False
+
+                try:
+                    from application.core.settings import settings
+                    compression_enabled = settings.ENABLE_CONVERSATION_COMPRESSION
+                except Exception:
+                    compression_enabled = False
+
+                if compression_enabled:
+                    compression_attempted = True
+                    try:
+                        logger.info(
+                            f"Context limit reached with {len(tool_calls) - i} remaining tool calls. "
+                            f"Attempting mid-execution compression..."
+                        )
+
+                        # Trigger mid-execution compression (DB-backed if available, otherwise in-memory)
+                        compression_successful, rebuilt_messages = self._perform_mid_execution_compression(
+                            agent, updated_messages
+                        )
+
+                        if compression_successful and rebuilt_messages is not None:
+                            # Update the messages list with rebuilt compressed version
+                            updated_messages = rebuilt_messages
+
+                            # Yield compression success message
+                            yield {
+                                "type": "info",
+                                "data": {
+                                    "message": "Context window limit reached. Compressed conversation history to continue processing."
+                                }
+                            }
+
+                            logger.info(
+                                f"Mid-execution compression successful. Continuing with {len(tool_calls) - i} remaining tool calls."
+                            )
+                            # Proceed to execute the current tool call with the reduced context
+                        else:
+                            logger.warning("Mid-execution compression attempted but failed. Skipping remaining tools.")
+                    except Exception as e:
+                        logger.error(f"Error during mid-execution compression: {str(e)}", exc_info=True)
+                        compression_attempted = True
+                        compression_successful = False
+
+                # If compression wasn't attempted or failed, skip remaining tools
+                if not compression_successful:
+                    if i == 0:
+                        # Special case: limit reached before executing any tools
+                        # This can happen when previous tool responses pushed context over limit
+                        if compression_attempted:
+                            logger.warning(
+                                f"Context limit reached before executing any tools. "
+                                f"Compression attempted but failed. "
+                                f"Skipping all {len(tool_calls)} pending tool call(s). "
+                                f"This typically occurs when previous tool responses contained large amounts of data."
+                            )
+                        else:
+                            logger.warning(
+                                f"Context limit reached before executing any tools. "
+                                f"Skipping all {len(tool_calls)} pending tool call(s). "
+                                f"This typically occurs when previous tool responses contained large amounts of data. "
+                                f"Consider enabling compression or using a model with larger context window."
+                            )
+                    else:
+                        # Normal case: executed some tools, now stopping
+                        tool_word = "tool call" if i == 1 else "tool calls"
+                        remaining = len(tool_calls) - i
+                        remaining_word = "tool call" if remaining == 1 else "tool calls"
+                        if compression_attempted:
+                            logger.warning(
+                                f"Context limit reached after executing {i} {tool_word}. "
+                                f"Compression attempted but failed. "
+                                f"Skipping remaining {remaining} {remaining_word}."
+                            )
+                        else:
+                            logger.warning(
+                                f"Context limit reached after executing {i} {tool_word}. "
+                                f"Skipping remaining {remaining} {remaining_word}. "
+                                f"Consider enabling compression or using a model with larger context window."
+                            )
+
+                    # Mark remaining tools as skipped
+                    for remaining_call in tool_calls[i:]:
+                        skip_message = {
+                            "type": "tool_call",
+                            "data": {
+                                "tool_name": "system",
+                                "call_id": remaining_call.id,
+                                "action_name": remaining_call.name,
+                                "arguments": {},
+                                "result": "Skipped: Context limit reached. Too many tool calls in conversation.",
+                                "status": "skipped"
+                            }
+                        }
+                        yield skip_message
+
+                    # Set flag on agent
+                    agent.context_limit_reached = True
+                    break
             try:
                 self.tool_calls.append(call)
                 tool_executor_gen = agent._execute_tool_action(tools_dict, call)
@@ -205,21 +710,26 @@ class LLMHandler(ABC):
                     except StopIteration as e:
                         tool_response, call_id = e.value
                         break
+                    
+                function_call_content = {
+                    "function_call": {
+                        "name": call.name,
+                        "args": call.arguments,
+                        "call_id": call_id,
+                    }
+                }
+                # Include thought_signature for Google Gemini 3 models
+                # It should be at the same level as function_call, not inside it
+                if call.thought_signature:
+                    function_call_content["thought_signature"] = call.thought_signature
                 updated_messages.append(
                     {
                         "role": "assistant",
-                        "content": [
-                            {
-                                "function_call": {
-                                    "name": call.name,
-                                    "args": call.arguments,
-                                    "call_id": call_id,
-                                }
-                            }
-                        ],
+                        "content": [function_call_content],
                     }
                 )
 
+
                 updated_messages.append(self.create_tool_message(call, tool_response))
             except Exception as e:
                 logger.error(f"Error executing tool: {str(e)}", exc_info=True)
@@ -282,7 +792,7 @@ class LLMHandler(ABC):
                     messages = e.value
                     break
             response = agent.llm.gen(
-                model=agent.gpt_model, messages=messages, tools=agent.tools
+                model=agent.model_id, messages=messages, tools=agent.tools
             )
             parsed = self.parse_response(response)
             self.llm_calls.append(build_stack_data(agent.llm))
@@ -324,6 +834,9 @@ class LLMHandler(ABC):
                             existing.name = call.name
                         if call.arguments:
                             existing.arguments += call.arguments
+                        # Preserve thought_signature for Google Gemini 3 models
+                        if call.thought_signature:
+                            existing.thought_signature = call.thought_signature
             if parsed.finish_reason == "tool_calls":
                 tool_handler_gen = self.handle_tool_calls(
                     agent, list(tool_calls.values()), tools_dict, messages
@@ -336,8 +849,21 @@ class LLMHandler(ABC):
                         break
                 tool_calls = {}
 
+                # Check if context limit was reached during tool execution
+                if hasattr(agent, 'context_limit_reached') and agent.context_limit_reached:
+                    # Add system message warning about context limit
+                    messages.append({
+                        "role": "system",
+                        "content": (
+                            "WARNING: Context window limit has been reached. "
+                            "Please provide a final response to the user without making additional tool calls. "
+                            "Summarize the work completed so far."
+                        )
+                    })
+                    logger.info("Context limit reached - instructing agent to wrap up")
+
                 response = agent.llm.gen_stream(
-                    model=agent.gpt_model, messages=messages, tools=agent.tools
+                    model=agent.model_id, messages=messages, tools=agent.tools if not agent.context_limit_reached else None
                 )
                 self.llm_calls.append(build_stack_data(agent.llm))
 

application/llm/handlers/google.py

@@ -19,15 +19,20 @@ class GoogleLLMHandler(LLMHandler):
             )
         if hasattr(response, "candidates"):
             parts = response.candidates[0].content.parts if response.candidates else []
-            tool_calls = [
-                ToolCall(
-                    id=str(uuid.uuid4()),
-                    name=part.function_call.name,
-                    arguments=part.function_call.args,
-                )
-                for part in parts
-                if hasattr(part, "function_call") and part.function_call is not None
-            ]
+            tool_calls = []
+            for idx, part in enumerate(parts):
+                if hasattr(part, "function_call") and part.function_call is not None:
+                    has_sig = hasattr(part, "thought_signature") and part.thought_signature is not None
+                    thought_sig = part.thought_signature if has_sig else None
+                    tool_calls.append(
+                        ToolCall(
+                            id=str(uuid.uuid4()),
+                            name=part.function_call.name,
+                            arguments=part.function_call.args,
+                            index=idx,
+                            thought_signature=thought_sig,
+                        )
+                    )
 
             content = " ".join(
                 part.text
@@ -41,13 +46,17 @@ class GoogleLLMHandler(LLMHandler):
                 raw_response=response,
             )
         else:
+            # This branch handles individual Part objects from streaming responses
             tool_calls = []
-            if hasattr(response, "function_call"):
+            if hasattr(response, "function_call") and response.function_call is not None:
+                has_sig = hasattr(response, "thought_signature") and response.thought_signature is not None
+                thought_sig = response.thought_signature if has_sig else None
                 tool_calls.append(
                     ToolCall(
                         id=str(uuid.uuid4()),
                         name=response.function_call.name,
                         arguments=response.function_call.args,
+                        thought_signature=thought_sig,
                     )
                 )
             return LLMResponse(

application/llm/llm_creator.py

@@ -1,13 +1,17 @@
-from application.llm.groq import GroqLLM
-from application.llm.openai import OpenAILLM, AzureOpenAILLM
-from application.llm.sagemaker import SagemakerAPILLM
-from application.llm.huggingface import HuggingFaceLLM
-from application.llm.llama_cpp import LlamaCpp
+import logging
+
 from application.llm.anthropic import AnthropicLLM
 from application.llm.docsgpt_provider import DocsGPTAPILLM
-from application.llm.premai import PremAILLM
 from application.llm.google_ai import GoogleLLM
+from application.llm.groq import GroqLLM
+from application.llm.huggingface import HuggingFaceLLM
+from application.llm.llama_cpp import LlamaCpp
 from application.llm.novita import NovitaLLM
+from application.llm.openai import AzureOpenAILLM, OpenAILLM
+from application.llm.premai import PremAILLM
+from application.llm.sagemaker import SagemakerAPILLM
+
+logger = logging.getLogger(__name__)
 
 
 class LLMCreator:
@@ -26,10 +30,26 @@ class LLMCreator:
     }
 
     @classmethod
-    def create_llm(cls, type, api_key, user_api_key, decoded_token, *args, **kwargs):
+    def create_llm(
+        cls, type, api_key, user_api_key, decoded_token, model_id=None, *args, **kwargs
+    ):
+        from application.core.model_utils import get_base_url_for_model
+
         llm_class = cls.llms.get(type.lower())
         if not llm_class:
             raise ValueError(f"No LLM class found for type {type}")
+
+        # Extract base_url from model configuration if model_id is provided
+        base_url = None
+        if model_id:
+            base_url = get_base_url_for_model(model_id)
+
         return llm_class(
-            api_key, user_api_key, decoded_token=decoded_token, *args, **kwargs
+            api_key,
+            user_api_key,
+            decoded_token=decoded_token,
+            model_id=model_id,
+            base_url=base_url,
+            *args,
+            **kwargs,
         )

application/llm/openai.py

@@ -2,6 +2,8 @@ import base64
 import json
 import logging
 
+from openai import OpenAI
+
 from application.core.settings import settings
 from application.llm.base import BaseLLM
 from application.storage.storage_creator import StorageCreator
@@ -9,20 +11,25 @@ from application.storage.storage_creator import StorageCreator
 
 class OpenAILLM(BaseLLM):
 
-    def __init__(self, api_key=None, user_api_key=None, *args, **kwargs):
-        from openai import OpenAI
+    def __init__(self, api_key=None, user_api_key=None, base_url=None, *args, **kwargs):
 
         super().__init__(*args, **kwargs)
-        if (
+        self.api_key = api_key or settings.OPENAI_API_KEY or settings.API_KEY
+        self.user_api_key = user_api_key
+
+        # Priority: 1) Parameter base_url, 2) Settings OPENAI_BASE_URL, 3) Default
+        effective_base_url = None
+        if base_url and isinstance(base_url, str) and base_url.strip():
+            effective_base_url = base_url
+        elif (
             isinstance(settings.OPENAI_BASE_URL, str)
             and settings.OPENAI_BASE_URL.strip()
         ):
-            self.client = OpenAI(api_key=api_key, base_url=settings.OPENAI_BASE_URL)
+            effective_base_url = settings.OPENAI_BASE_URL
         else:
-            DEFAULT_OPENAI_API_BASE = "https://api.openai.com/v1"
-            self.client = OpenAI(api_key=api_key, base_url=DEFAULT_OPENAI_API_BASE)
-        self.api_key = api_key
-        self.user_api_key = user_api_key
+            effective_base_url = "https://api.openai.com/v1"
+
+        self.client = OpenAI(api_key=self.api_key, base_url=effective_base_url)
         self.storage = StorageCreator.get_storage()
 
     def _clean_messages_openai(self, messages):
@@ -33,7 +40,6 @@ class OpenAILLM(BaseLLM):
 
             if role == "model":
                 role = "assistant"
-
             if role and content is not None:
                 if isinstance(content, str):
                     cleaned_messages.append({"role": role, "content": content})
@@ -44,14 +50,15 @@ class OpenAILLM(BaseLLM):
                                 {"role": role, "content": item["text"]}
                             )
                         elif "function_call" in item:
+                            cleaned_args = self._remove_null_values(
+                                item["function_call"]["args"]
+                            )
                             tool_call = {
                                 "id": item["function_call"]["call_id"],
                                 "type": "function",
                                 "function": {
                                     "name": item["function_call"]["name"],
-                                    "arguments": json.dumps(
-                                        item["function_call"]["args"]
-                                    ),
+                                    "arguments": json.dumps(cleaned_args),
                                 },
                             }
                             cleaned_messages.append(
@@ -106,7 +113,6 @@ class OpenAILLM(BaseLLM):
                             )
                 else:
                     raise ValueError(f"Unexpected content type: {type(content)}")
-
         return cleaned_messages
 
     def _raw_gen(
@@ -122,6 +128,10 @@ class OpenAILLM(BaseLLM):
     ):
         messages = self._clean_messages_openai(messages)
 
+        # Convert max_tokens to max_completion_tokens for newer models
+        if "max_tokens" in kwargs:
+            kwargs["max_completion_tokens"] = kwargs.pop("max_tokens")
+
         request_params = {
             "model": model,
             "messages": messages,
@@ -131,10 +141,8 @@ class OpenAILLM(BaseLLM):
 
         if tools:
             request_params["tools"] = tools
-
         if response_format:
             request_params["response_format"] = response_format
-
         response = self.client.chat.completions.create(**request_params)
 
         if tools:
@@ -155,6 +163,10 @@ class OpenAILLM(BaseLLM):
     ):
         messages = self._clean_messages_openai(messages)
 
+        # Convert max_tokens to max_completion_tokens for newer models
+        if "max_tokens" in kwargs:
+            kwargs["max_completion_tokens"] = kwargs.pop("max_tokens")
+
         request_params = {
             "model": model,
             "messages": messages,
@@ -164,10 +176,8 @@ class OpenAILLM(BaseLLM):
 
         if tools:
             request_params["tools"] = tools
-
         if response_format:
             request_params["response_format"] = response_format
-
         response = self.client.chat.completions.create(**request_params)
 
         try:
@@ -181,7 +191,7 @@ class OpenAILLM(BaseLLM):
                 elif len(line.choices) > 0:
                     yield line.choices[0]
         finally:
-            if hasattr(response, 'close'):
+            if hasattr(response, "close"):
                 response.close()
 
     def _supports_tools(self):
@@ -193,7 +203,6 @@ class OpenAILLM(BaseLLM):
     def prepare_structured_output_format(self, json_schema):
         if not json_schema:
             return None
-
         try:
 
             def add_additional_properties_false(schema_obj):
@@ -203,11 +212,11 @@ class OpenAILLM(BaseLLM):
                     if schema_copy.get("type") == "object":
                         schema_copy["additionalProperties"] = False
                         # Ensure 'required' includes all properties for OpenAI strict mode
+
                         if "properties" in schema_copy:
                             schema_copy["required"] = list(
                                 schema_copy["properties"].keys()
                             )
-
                     for key, value in schema_copy.items():
                         if key == "properties" and isinstance(value, dict):
                             schema_copy[key] = {
@@ -223,7 +232,6 @@ class OpenAILLM(BaseLLM):
                                 add_additional_properties_false(sub_schema)
                                 for sub_schema in value
                             ]
-
                     return schema_copy
                 return schema_obj
 
@@ -242,7 +250,6 @@ class OpenAILLM(BaseLLM):
             }
 
             return result
-
         except Exception as e:
             logging.error(f"Error preparing structured output format: {e}")
             return None
@@ -276,21 +283,19 @@ class OpenAILLM(BaseLLM):
         """
         if not attachments:
             return messages
-
         prepared_messages = messages.copy()
 
         # Find the user message to attach file_id to the last one
+
         user_message_index = None
         for i in range(len(prepared_messages) - 1, -1, -1):
             if prepared_messages[i].get("role") == "user":
                 user_message_index = i
                 break
-
         if user_message_index is None:
             user_message = {"role": "user", "content": []}
             prepared_messages.append(user_message)
             user_message_index = len(prepared_messages) - 1
-
         if isinstance(prepared_messages[user_message_index].get("content"), str):
             text_content = prepared_messages[user_message_index]["content"]
             prepared_messages[user_message_index]["content"] = [
@@ -298,7 +303,6 @@ class OpenAILLM(BaseLLM):
             ]
         elif not isinstance(prepared_messages[user_message_index].get("content"), list):
             prepared_messages[user_message_index]["content"] = []
-
         for attachment in attachments:
             mime_type = attachment.get("mime_type")
 
@@ -325,6 +329,7 @@ class OpenAILLM(BaseLLM):
                             }
                         )
             # Handle PDFs using the file API
+
             elif mime_type == "application/pdf":
                 try:
                     file_id = self._upload_file_to_openai(attachment)
@@ -340,7 +345,6 @@ class OpenAILLM(BaseLLM):
                                 "text": f"File content:\n\n{attachment['content']}",
                             }
                         )
-
         return prepared_messages
 
     def _get_base64_image(self, attachment):
@@ -356,7 +360,6 @@ class OpenAILLM(BaseLLM):
         file_path = attachment.get("path")
         if not file_path:
             raise ValueError("No file path provided in attachment")
-
         try:
             with self.storage.get_file(file_path) as image_file:
                 return base64.b64encode(image_file.read()).decode("utf-8")
@@ -380,12 +383,10 @@ class OpenAILLM(BaseLLM):
 
         if "openai_file_id" in attachment:
             return attachment["openai_file_id"]
-
         file_path = attachment.get("path")
 
         if not self.storage.file_exists(file_path):
             raise FileNotFoundError(f"File not found: {file_path}")
-
         try:
             file_id = self.storage.process_file(
                 file_path,
@@ -403,7 +404,6 @@ class OpenAILLM(BaseLLM):
                 attachments_collection.update_one(
                     {"_id": attachment["_id"]}, {"$set": {"openai_file_id": file_id}}
                 )
-
             return file_id
         except Exception as e:
             logging.error(f"Error uploading file to OpenAI: {e}", exc_info=True)

application/parser/embedding_pipeline.py

@@ -1,5 +1,6 @@
 import os
 import logging
+from typing import List, Any
 from retry import retry
 from tqdm import tqdm
 from application.core.settings import settings
@@ -22,13 +23,16 @@ def sanitize_content(content: str) -> str:
 
 
 @retry(tries=10, delay=60)
-def add_text_to_store_with_retry(store, doc, source_id):
-    """
-    Add a document's text and metadata to the vector store with retry logic.
+def add_text_to_store_with_retry(store: Any, doc: Any, source_id: str) -> None:
+    """Add a document's text and metadata to the vector store with retry logic.
+    
     Args:
         store: The vector store object.
         doc: The document to be added.
         source_id: Unique identifier for the source.
+        
+    Raises:
+        Exception: If document addition fails after all retry attempts.
     """
     try:
         # Sanitize content to remove NUL characters that cause ingestion failures
@@ -41,18 +45,21 @@ def add_text_to_store_with_retry(store, doc, source_id):
         raise
 
 
-def embed_and_store_documents(docs, folder_name, source_id, task_status):
-    """
-    Embeds documents and stores them in a vector store.
+def embed_and_store_documents(docs: List[Any], folder_name: str, source_id: str, task_status: Any) -> None:
+    """Embeds documents and stores them in a vector store.
 
     Args:
-        docs (list): List of documents to be embedded and stored.
-        folder_name (str): Directory to save the vector store.
-        source_id (str): Unique identifier for the source.
+        docs: List of documents to be embedded and stored.
+        folder_name: Directory to save the vector store.
+        source_id: Unique identifier for the source.
         task_status: Task state manager for progress updates.
 
     Returns:
         None
+        
+    Raises:
+        OSError: If unable to create folder or save vector store.
+        Exception: If vector store creation or document embedding fails.
     """
     # Ensure the folder exists
     if not os.path.exists(folder_name):
@@ -95,10 +102,21 @@ def embed_and_store_documents(docs, folder_name, source_id, task_status):
         except Exception as e:
             logging.error(f"Error embedding document {idx}: {e}", exc_info=True)
             logging.info(f"Saving progress at document {idx} out of {total_docs}")
-            store.save_local(folder_name)
+            try:
+                store.save_local(folder_name)
+                logging.info("Progress saved successfully")
+            except Exception as save_error:
+                logging.error(f"CRITICAL: Failed to save progress: {save_error}", exc_info=True)
+                # Continue without breaking to attempt final save
             break
 
     # Save the vector store
     if settings.VECTOR_STORE == "faiss":
-        store.save_local(folder_name)
-    logging.info("Vector store saved successfully.")
+        try:
+            store.save_local(folder_name)
+            logging.info("Vector store saved successfully.")
+        except Exception as e:
+            logging.error(f"CRITICAL: Failed to save final vector store: {e}", exc_info=True)
+            raise OSError(f"Unable to save vector store to {folder_name}: {e}") from e
+    else:
+        logging.info("Vector store saved successfully.")

application/prompts/compression/v1.0.txt

@@ -0,0 +1,35 @@
+Your task is to create a detailed summary of the conversation so far, paying close attention to the user's explicit requests and your previous actions.
+
+This summary should be thorough in capturing technical details, code patterns, and architectural decisions that would be essential for continuing work without losing context.
+
+Before providing your final summary, wrap your analysis in <analysis> tags to organize your thoughts and ensure you've covered all necessary points. In your analysis process:
+
+1. Chronologically analyze each message, tool call and section of the conversation. For each section thoroughly identify:
+   - The user's explicit requests and intents
+   - Your approach to addressing the user's requests
+   - Key decisions, concepts and patterns
+   - Specific details like if applicable:
+     - file names
+     - full code snippets
+     - function signatures
+     - file edits
+   - Errors that you ran into and how you fixed them
+   - Pay special attention to specific user feedback that you received, especially if the user told you to do something differently.
+
+2. Double-check for accuracy and completeness, addressing each required element thoroughly.
+
+Your summary should include the following sections:
+
+1. Primary Request and Intent: Capture all of the user's explicit requests and intents in detail
+2. Key Concepts: List all important concepts discussed.
+3. Files and Code Sections: Enumerate specific files and code sections examined, modified, or created. Pay special attention to the most recent messages and include full code snippets where applicable and include a summary of why this file read or edit is important.
+4. Errors and fixes: List all errors that you ran into, and how you fixed them. Pay special attention to specific user feedback that you received, especially if the user told you to do something differently.
+5. Problem Solving: Document problems solved and any ongoing troubleshooting efforts.
+6. All user messages: List ALL user messages that are not tool results. These are critical for understanding the users' feedback and changing intent.
+7. Tool Calls: List ALL tool calls made, including their inputs relevant parts of the outputs.
+8. Pending Tasks: Outline any pending tasks that you have explicitly been asked to work on.
+9. Current Work: Describe in detail precisely what was being worked on immediately before this summary request, paying special attention to the most recent messages from both user and assistant. Include file names and code snippets where applicable.
+10. Optional Next Step: List the next step that you will take that is related to the most recent work you were doing. IMPORTANT: ensure that this step is DIRECTLY in line with the user's most recent explicit requests, and the task you were working on immediately before this summary request. If your last task was concluded, then only list next steps if they are explicitly in line with the users request. Do not start on tangential requests or really old requests that were already completed without confirming with the user first.
+If there is a next step, include direct quotes from the most recent conversation showing exactly what task you were working on and where you left off. This should be verbatim to ensure there's no drift in task interpretation.
+
+Please provide your summary based on the conversation and tools used so far, following this structure and ensuring precision and thoroughness in your response.

application/requirements.txt

@@ -15,7 +15,7 @@ Flask==3.1.1
 faiss-cpu==1.9.0.post1
 fastmcp==2.11.0
 flask-restx==1.3.0
-google-genai==1.3.0
+google-genai==1.49.0
 google-api-python-client==2.179.0
 google-auth-httplib2==0.2.0
 google-auth-oauthlib==1.2.2

application/retriever/base.py

@@ -8,7 +8,3 @@ class BaseRetriever(ABC):
     @abstractmethod
     def search(self, *args, **kwargs):
         pass
-
-    @abstractmethod
-    def get_params(self):
-        pass

application/retriever/classic_rag.py

@@ -4,7 +4,7 @@ import os
 from application.core.settings import settings
 from application.llm.llm_creator import LLMCreator
 from application.retriever.base import BaseRetriever
-
+from application.utils import num_tokens_from_string
 from application.vectorstore.vector_creator import VectorCreator
 
 
@@ -15,14 +15,13 @@ class ClassicRAG(BaseRetriever):
         chat_history=None,
         prompt="",
         chunks=2,
-        token_limit=150,
-        gpt_model="docsgpt",
+        doc_token_limit=50000,
+        model_id="docsgpt-local",
         user_api_key=None,
         llm_name=settings.LLM_PROVIDER,
         api_key=settings.API_KEY,
         decoded_token=None,
     ):
-        """Initialize ClassicRAG retriever with vectorstore sources and LLM configuration"""
         self.original_question = source.get("question", "")
         self.chat_history = chat_history if chat_history is not None else []
         self.prompt = prompt
@@ -41,17 +40,8 @@ class ClassicRAG(BaseRetriever):
             f"ClassicRAG initialized with chunks={self.chunks}, user_api_key={user_identifier}, "
             f"sources={'active_docs' in source and source['active_docs'] is not None}"
         )
-        self.gpt_model = gpt_model
-        self.token_limit = (
-            token_limit
-            if token_limit
-            < settings.LLM_TOKEN_LIMITS.get(
-                self.gpt_model, settings.DEFAULT_MAX_HISTORY
-            )
-            else settings.LLM_TOKEN_LIMITS.get(
-                self.gpt_model, settings.DEFAULT_MAX_HISTORY
-            )
-        )
+        self.model_id = model_id
+        self.doc_token_limit = doc_token_limit
         self.user_api_key = user_api_key
         self.llm_name = llm_name
         self.api_key = api_key
@@ -110,7 +100,7 @@ class ClassicRAG(BaseRetriever):
         ]
 
         try:
-            rephrased_query = self.llm.gen(model=self.gpt_model, messages=messages)
+            rephrased_query = self.llm.gen(model=self.model_id, messages=messages)
             print(f"Rephrased query: {rephrased_query}")
             return rephrased_query if rephrased_query else self.original_question
         except Exception as e:
@@ -118,21 +108,17 @@ class ClassicRAG(BaseRetriever):
             return self.original_question
 
     def _get_data(self):
-        """Retrieve relevant documents from configured vectorstores"""
         if self.chunks == 0 or not self.vectorstores:
             logging.info(
                 f"ClassicRAG._get_data: Skipping retrieval - chunks={self.chunks}, "
                 f"vectorstores_count={len(self.vectorstores) if self.vectorstores else 0}"
             )
             return []
+
         all_docs = []
         chunks_per_source = max(1, self.chunks // len(self.vectorstores))
-
-        logging.info(
-            f"ClassicRAG._get_data: Starting retrieval with chunks={self.chunks}, "
-            f"vectorstores={self.vectorstores}, chunks_per_source={chunks_per_source}, "
-            f"query='{self.question[:50]}...'"
-        )
+        token_budget = max(int(self.doc_token_limit * 0.9), 100)
+        cumulative_tokens = 0
 
         for vectorstore_id in self.vectorstores:
             if vectorstore_id:
@@ -140,15 +126,21 @@ class ClassicRAG(BaseRetriever):
                     docsearch = VectorCreator.create_vectorstore(
                         settings.VECTOR_STORE, vectorstore_id, settings.EMBEDDINGS_KEY
                     )
-                    docs_temp = docsearch.search(self.question, k=chunks_per_source)
+                    docs_temp = docsearch.search(
+                        self.question, k=max(chunks_per_source * 2, 20)
+                    )
 
                     for doc in docs_temp:
+                        if cumulative_tokens >= token_budget:
+                            break
+
                         if hasattr(doc, "page_content") and hasattr(doc, "metadata"):
                             page_content = doc.page_content
                             metadata = doc.metadata
                         else:
                             page_content = doc.get("text", doc.get("page_content", ""))
                             metadata = doc.get("metadata", {})
+
                         title = metadata.get(
                             "title", metadata.get("post_title", page_content)
                         )
@@ -168,23 +160,35 @@ class ClassicRAG(BaseRetriever):
                         if not filename:
                             filename = title
                         source_path = metadata.get("source") or vectorstore_id
-                        all_docs.append(
-                            {
-                                "title": title,
-                                "text": page_content,
-                                "source": source_path,
-                                "filename": filename,
-                            }
-                        )
+
+                        doc_text_with_header = f"{filename}\n{page_content}"
+                        doc_tokens = num_tokens_from_string(doc_text_with_header)
+
+                        if cumulative_tokens + doc_tokens < token_budget:
+                            all_docs.append(
+                                {
+                                    "title": title,
+                                    "text": page_content,
+                                    "source": source_path,
+                                    "filename": filename,
+                                }
+                            )
+                            cumulative_tokens += doc_tokens
+
+                    if cumulative_tokens >= token_budget:
+                        break
+
                 except Exception as e:
                     logging.error(
                         f"Error searching vectorstore {vectorstore_id}: {e}",
                         exc_info=True,
                     )
                     continue
+
         logging.info(
             f"ClassicRAG._get_data: Retrieval complete - retrieved {len(all_docs)} documents "
-            f"(requested chunks={self.chunks}, chunks_per_source={chunks_per_source})"
+            f"(requested chunks={self.chunks}, chunks_per_source={chunks_per_source}, "
+            f"cumulative_tokens={cumulative_tokens}/{token_budget})"
         )
         return all_docs
 
@@ -194,15 +198,3 @@ class ClassicRAG(BaseRetriever):
             self.original_question = query
             self.question = self._rephrase_query()
         return self._get_data()
-
-    def get_params(self):
-        """Return current retriever configuration parameters"""
-        return {
-            "question": self.original_question,
-            "rephrased_question": self.question,
-            "sources": self.vectorstores,
-            "chunks": self.chunks,
-            "token_limit": self.token_limit,
-            "gpt_model": self.gpt_model,
-            "user_api_key": self.user_api_key,
-        }

application/templates/namespaces.py

@@ -0,0 +1,190 @@
+import logging
+import uuid
+from abc import ABC, abstractmethod
+from datetime import datetime, timezone
+from typing import Any, Dict, Optional
+
+logger = logging.getLogger(__name__)
+
+
+class NamespaceBuilder(ABC):
+    """Base class for building template context namespaces"""
+
+    @abstractmethod
+    def build(self, **kwargs) -> Dict[str, Any]:
+        """Build namespace context dictionary"""
+        pass
+
+    @property
+    @abstractmethod
+    def namespace_name(self) -> str:
+        """Name of this namespace for template access"""
+        pass
+
+
+class SystemNamespace(NamespaceBuilder):
+    """System metadata namespace: {{ system.* }}"""
+
+    @property
+    def namespace_name(self) -> str:
+        return "system"
+
+    def build(
+        self, request_id: Optional[str] = None, user_id: Optional[str] = None, **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Build system context with metadata.
+
+        Args:
+            request_id: Unique request identifier
+            user_id: Current user identifier
+
+        Returns:
+            Dictionary with system variables
+        """
+        now = datetime.now(timezone.utc)
+
+        return {
+            "date": now.strftime("%Y-%m-%d"),
+            "time": now.strftime("%H:%M:%S"),
+            "timestamp": now.isoformat(),
+            "request_id": request_id or str(uuid.uuid4()),
+            "user_id": user_id,
+        }
+
+
+class PassthroughNamespace(NamespaceBuilder):
+    """Request parameters namespace: {{ passthrough.* }}"""
+
+    @property
+    def namespace_name(self) -> str:
+        return "passthrough"
+
+    def build(
+        self, passthrough_data: Optional[Dict[str, Any]] = None, **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Build passthrough context from request parameters.
+
+        Args:
+            passthrough_data: Dictionary of parameters from web request
+
+        Returns:
+            Dictionary with passthrough variables
+        """
+        if not passthrough_data:
+            return {}
+        safe_data = {}
+        for key, value in passthrough_data.items():
+            if isinstance(value, (str, int, float, bool, type(None))):
+                safe_data[key] = value
+            else:
+                logger.warning(
+                    f"Skipping non-serializable passthrough value for key '{key}': {type(value)}"
+                )
+        return safe_data
+
+
+class SourceNamespace(NamespaceBuilder):
+    """RAG source documents namespace: {{ source.* }}"""
+
+    @property
+    def namespace_name(self) -> str:
+        return "source"
+
+    def build(
+        self, docs: Optional[list] = None, docs_together: Optional[str] = None, **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Build source context from RAG retrieval results.
+
+        Args:
+            docs: List of retrieved documents
+            docs_together: Concatenated document content (for backward compatibility)
+
+        Returns:
+            Dictionary with source variables
+        """
+        context = {}
+
+        if docs:
+            context["documents"] = docs
+            context["count"] = len(docs)
+        if docs_together:
+            context["docs_together"] = docs_together  # Add docs_together for custom templates
+            context["content"] = docs_together
+            context["summaries"] = docs_together
+        return context
+
+
+class ToolsNamespace(NamespaceBuilder):
+    """Pre-executed tools namespace: {{ tools.* }}"""
+
+    @property
+    def namespace_name(self) -> str:
+        return "tools"
+
+    def build(
+        self, tools_data: Optional[Dict[str, Any]] = None, **kwargs
+    ) -> Dict[str, Any]:
+        """
+        Build tools context with pre-executed tool results.
+
+        Args:
+            tools_data: Dictionary of pre-fetched tool results organized by tool name
+                       e.g., {"memory": {"notes": "content", "tasks": "list"}}
+
+        Returns:
+            Dictionary with tool results organized by tool name
+        """
+        if not tools_data:
+            return {}
+
+        safe_data = {}
+        for tool_name, tool_result in tools_data.items():
+            if isinstance(tool_result, (str, dict, list, int, float, bool, type(None))):
+                safe_data[tool_name] = tool_result
+            else:
+                logger.warning(
+                    f"Skipping non-serializable tool result for '{tool_name}': {type(tool_result)}"
+                )
+        return safe_data
+
+
+class NamespaceManager:
+    """Manages all namespace builders and context assembly"""
+
+    def __init__(self):
+        self._builders = {
+            "system": SystemNamespace(),
+            "passthrough": PassthroughNamespace(),
+            "source": SourceNamespace(),
+            "tools": ToolsNamespace(),
+        }
+
+    def build_context(self, **kwargs) -> Dict[str, Any]:
+        """
+        Build complete template context from all namespaces.
+
+        Args:
+            **kwargs: Parameters to pass to namespace builders
+
+        Returns:
+            Complete context dictionary for template rendering
+        """
+        context = {}
+
+        for namespace_name, builder in self._builders.items():
+            try:
+                namespace_context = builder.build(**kwargs)
+                # Always include namespace, even if empty, to prevent undefined errors
+                context[namespace_name] = namespace_context if namespace_context else {}
+            except Exception as e:
+                logger.error(f"Failed to build {namespace_name} namespace: {str(e)}")
+                # Include empty namespace on error to prevent template failures
+                context[namespace_name] = {}
+        return context
+
+    def get_builder(self, namespace_name: str) -> Optional[NamespaceBuilder]:
+        """Get specific namespace builder"""
+        return self._builders.get(namespace_name)

application/templates/template_engine.py

@@ -0,0 +1,161 @@
+import logging
+from typing import Any, Dict, List, Optional, Set
+
+from jinja2 import (
+    ChainableUndefined,
+    Environment,
+    nodes,
+    select_autoescape,
+    TemplateSyntaxError,
+)
+from jinja2.exceptions import UndefinedError
+
+logger = logging.getLogger(__name__)
+
+
+class TemplateRenderError(Exception):
+    """Raised when template rendering fails"""
+
+    pass
+
+
+class TemplateEngine:
+    """Jinja2-based template engine for dynamic prompt rendering"""
+
+    def __init__(self):
+        self._env = Environment(
+            undefined=ChainableUndefined,
+            trim_blocks=True,
+            lstrip_blocks=True,
+            autoescape=select_autoescape(default_for_string=True, default=True),
+        )
+
+    def render(self, template_content: str, context: Dict[str, Any]) -> str:
+        """
+        Render template with provided context.
+
+        Args:
+            template_content: Raw template string with Jinja2 syntax
+            context: Dictionary of variables to inject into template
+
+        Returns:
+            Rendered template string
+
+        Raises:
+            TemplateRenderError: If template syntax is invalid or variables undefined
+        """
+        if not template_content:
+            return ""
+        try:
+            template = self._env.from_string(template_content)
+            return template.render(**context)
+        except TemplateSyntaxError as e:
+            error_msg = f"Template syntax error at line {e.lineno}: {e.message}"
+            logger.error(error_msg)
+            raise TemplateRenderError(error_msg) from e
+        except UndefinedError as e:
+            error_msg = f"Undefined variable in template: {e.message}"
+            logger.error(error_msg)
+            raise TemplateRenderError(error_msg) from e
+        except Exception as e:
+            error_msg = f"Template rendering failed: {str(e)}"
+            logger.error(error_msg)
+            raise TemplateRenderError(error_msg) from e
+
+    def validate_template(self, template_content: str) -> bool:
+        """
+        Validate template syntax without rendering.
+
+        Args:
+            template_content: Template string to validate
+
+        Returns:
+            True if template is syntactically valid
+        """
+        if not template_content:
+            return True
+        try:
+            self._env.from_string(template_content)
+            return True
+        except TemplateSyntaxError as e:
+            logger.debug(f"Template syntax invalid at line {e.lineno}: {e.message}")
+            return False
+        except Exception as e:
+            logger.debug(f"Template validation error: {type(e).__name__}: {str(e)}")
+            return False
+
+    def extract_variables(self, template_content: str) -> Set[str]:
+        """
+        Extract all variable names from template.
+
+        Args:
+            template_content: Template string to analyze
+
+        Returns:
+            Set of variable names found in template
+        """
+        if not template_content:
+            return set()
+        try:
+            ast = self._env.parse(template_content)
+            return set(self._env.get_template_module(ast).make_module().keys())
+        except TemplateSyntaxError as e:
+            logger.debug(f"Cannot extract variables - syntax error at line {e.lineno}")
+            return set()
+        except Exception as e:
+            logger.debug(f"Cannot extract variables: {type(e).__name__}")
+            return set()
+
+    def extract_tool_usages(
+        self, template_content: str
+    ) -> Dict[str, Set[Optional[str]]]:
+        """Extract tool and action references from a template"""
+        if not template_content:
+            return {}
+        try:
+            ast = self._env.parse(template_content)
+        except TemplateSyntaxError as e:
+            logger.debug(f"extract_tool_usages - syntax error at line {e.lineno}")
+            return {}
+        except Exception as e:
+            logger.debug(f"extract_tool_usages - parse error: {type(e).__name__}")
+            return {}
+
+        usages: Dict[str, Set[Optional[str]]] = {}
+
+        def record(path: List[str]) -> None:
+            if not path:
+                return
+            tool_name = path[0]
+            action_name = path[1] if len(path) > 1 else None
+            if not tool_name:
+                return
+            tool_entry = usages.setdefault(tool_name, set())
+            tool_entry.add(action_name)
+
+        for node in ast.find_all(nodes.Getattr):
+            path = []
+            current = node
+            while isinstance(current, nodes.Getattr):
+                path.append(current.attr)
+                current = current.node
+            if isinstance(current, nodes.Name) and current.name == "tools":
+                path.reverse()
+                record(path)
+
+        for node in ast.find_all(nodes.Getitem):
+            path = []
+            current = node
+            while isinstance(current, nodes.Getitem):
+                key = current.arg
+                if isinstance(key, nodes.Const) and isinstance(key.value, str):
+                    path.append(key.value)
+                else:
+                    path = []
+                    break
+                current = current.node
+            if path and isinstance(current, nodes.Name) and current.name == "tools":
+                path.reverse()
+                record(path)
+
+        return usages

application/utils.py

@@ -7,6 +7,8 @@ import tiktoken
 from flask import jsonify, make_response
 from werkzeug.utils import secure_filename
 
+from application.core.model_utils import get_token_limit
+
 from application.core.settings import settings
 
 
@@ -74,6 +76,15 @@ def count_tokens_docs(docs):
     return tokens
 
 
+def calculate_doc_token_budget(
+    model_id: str = "gpt-4o", history_token_limit: int = 2000
+) -> int:
+    total_context = get_token_limit(model_id)
+    reserved = sum(settings.RESERVED_TOKENS.values())
+    doc_budget = total_context - history_token_limit - reserved
+    return max(doc_budget, 1000)
+
+
 def get_missing_fields(data, required_fields):
     """Check for missing required fields. Returns list of missing field names."""
     return [field for field in required_fields if field not in data]
@@ -133,16 +144,13 @@ def get_hash(data):
     return hashlib.md5(data.encode(), usedforsecurity=False).hexdigest()
 
 
-def limit_chat_history(history, max_token_limit=None, gpt_model="docsgpt"):
+def limit_chat_history(history, max_token_limit=None, model_id="docsgpt-local"):
     """Limit chat history to fit within token limit."""
-    from application.core.settings import settings
-
+    model_token_limit = get_token_limit(model_id)
     max_token_limit = (
         max_token_limit
-        if max_token_limit
-        and max_token_limit
-        < settings.LLM_TOKEN_LIMITS.get(gpt_model, settings.DEFAULT_MAX_HISTORY)
-        else settings.LLM_TOKEN_LIMITS.get(gpt_model, settings.DEFAULT_MAX_HISTORY)
+        if max_token_limit and max_token_limit < model_token_limit
+        else model_token_limit
     )
 
     if not history:
@@ -187,3 +195,69 @@ def generate_image_url(image_path):
     else:
         base_url = getattr(settings, "API_URL", "http://localhost:7091")
         return f"{base_url}/api/images/{image_path}"
+
+
+def calculate_compression_threshold(
+    model_id: str, threshold_percentage: float = 0.8
+) -> int:
+    """
+    Calculate token threshold for triggering compression.
+
+    Args:
+        model_id: Model identifier
+        threshold_percentage: Percentage of context window (default 80%)
+
+    Returns:
+        Token count threshold
+    """
+    total_context = get_token_limit(model_id)
+    threshold = int(total_context * threshold_percentage)
+    return threshold
+
+
+def clean_text_for_tts(text: str) -> str:
+    """
+    clean text for Text-to-Speech processing.
+    """
+    # Handle code blocks and links
+
+    text = re.sub(r"```mermaid[\s\S]*?```", " flowchart, ", text)  ## ```mermaid...```
+    text = re.sub(r"```[\s\S]*?```", " code block, ", text)  ## ```code```
+    text = re.sub(r"\[([^\]]+)\]\([^\)]+\)", r"\1", text)  ## [text](url)
+    text = re.sub(r"!\[([^\]]*)\]\([^\)]+\)", "", text)  ## ![alt](url)
+
+    # Remove markdown formatting
+
+    text = re.sub(r"`([^`]+)`", r"\1", text)  ## `code`
+    text = re.sub(r"\{([^}]*)\}", r" \1 ", text)  ## {text}
+    text = re.sub(r"[{}]", " ", text)  ## unmatched {}
+    text = re.sub(r"\[([^\]]+)\]", r" \1 ", text)  ## [text]
+    text = re.sub(r"[\[\]]", " ", text)  ## unmatched []
+    text = re.sub(r"(\*\*|__)(.*?)\1", r"\2", text)  ## **bold** __bold__
+    text = re.sub(r"(\*|_)(.*?)\1", r"\2", text)  ## *italic* _italic_
+    text = re.sub(r"^#{1,6}\s+", "", text, flags=re.MULTILINE)  ## # headers
+    text = re.sub(r"^>\s+", "", text, flags=re.MULTILINE)  ## > blockquotes
+    text = re.sub(r"^[\s]*[-\*\+]\s+", "", text, flags=re.MULTILINE)  ## - * + lists
+    text = re.sub(r"^[\s]*\d+\.\s+", "", text, flags=re.MULTILINE)  ## 1. numbered lists
+    text = re.sub(
+        r"^[\*\-_]{3,}\s*$", "", text, flags=re.MULTILINE
+    )  ## --- *** ___ rules
+    text = re.sub(r"<[^>]*>", "", text)  ## <html> tags
+
+    # Remove non-ASCII (emojis, special Unicode)
+
+    text = re.sub(r"[^\x20-\x7E\n\r\t]", "", text)
+
+    # Replace special sequences
+
+    text = re.sub(r"-->", ", ", text)  ## -->
+    text = re.sub(r"<--", ", ", text)  ## <--
+    text = re.sub(r"=>", ", ", text)  ## =>
+    text = re.sub(r"::", " ", text)  ## ::
+
+    # Normalize whitespace
+
+    text = re.sub(r"\s+", " ", text)
+    text = text.strip()
+
+    return text

application/worker.py

@@ -146,6 +146,14 @@ def upload_index(full_path, file_data):
 
 def run_agent_logic(agent_config, input_data):
     try:
+        from application.core.model_utils import (
+            get_api_key_for_provider,
+            get_default_model_id,
+            get_provider_from_model_id,
+            validate_model_id,
+        )
+        from application.utils import calculate_doc_token_budget
+
         source = agent_config.get("source")
         retriever = agent_config.get("retriever", "classic")
         if isinstance(source, DBRef):
@@ -160,31 +168,62 @@ def run_agent_logic(agent_config, input_data):
         user_api_key = agent_config["key"]
         agent_type = agent_config.get("agent_type", "classic")
         decoded_token = {"sub": agent_config.get("user")}
+        json_schema = agent_config.get("json_schema")
         prompt = get_prompt(prompt_id, db["prompts"])
-        agent = AgentCreator.create_agent(
-            agent_type,
-            endpoint="webhook",
-            llm_name=settings.LLM_PROVIDER,
-            gpt_model=settings.LLM_NAME,
-            api_key=settings.API_KEY,
-            user_api_key=user_api_key,
-            prompt=prompt,
-            chat_history=[],
-            decoded_token=decoded_token,
-            attachments=[],
+
+        # Determine model_id: check agent's default_model_id, fallback to system default
+        agent_default_model = agent_config.get("default_model_id", "")
+        if agent_default_model and validate_model_id(agent_default_model):
+            model_id = agent_default_model
+        else:
+            model_id = get_default_model_id()
+
+        # Get provider and API key for the selected model
+        provider = get_provider_from_model_id(model_id) if model_id else settings.LLM_PROVIDER
+        system_api_key = get_api_key_for_provider(provider or settings.LLM_PROVIDER)
+
+        # Calculate proper doc_token_limit based on model's context window
+        history_token_limit = 2000  # Default for webhooks
+        doc_token_limit = calculate_doc_token_budget(
+            model_id=model_id, history_token_limit=history_token_limit
         )
+
         retriever = RetrieverCreator.create_retriever(
             retriever,
             source=source,
             chat_history=[],
             prompt=prompt,
             chunks=chunks,
-            token_limit=settings.DEFAULT_MAX_HISTORY,
-            gpt_model=settings.LLM_NAME,
+            doc_token_limit=doc_token_limit,
+            model_id=model_id,
             user_api_key=user_api_key,
             decoded_token=decoded_token,
         )
-        answer = agent.gen(query=input_data, retriever=retriever)
+
+        # Pre-fetch documents using the retriever
+        retrieved_docs = []
+        try:
+            docs = retriever.search(input_data)
+            if docs:
+                retrieved_docs = docs
+        except Exception as e:
+            logging.warning(f"Failed to retrieve documents: {e}")
+
+        agent = AgentCreator.create_agent(
+            agent_type,
+            endpoint="webhook",
+            llm_name=provider or settings.LLM_PROVIDER,
+            model_id=model_id,
+            api_key=system_api_key,
+            user_api_key=user_api_key,
+            prompt=prompt,
+            chat_history=[],
+            retrieved_docs=retrieved_docs,
+            decoded_token=decoded_token,
+            attachments=[],
+            json_schema=json_schema,
+        )
+        answer = agent.gen(query=input_data)
         response_full = ""
         thought = ""
         source_log_docs = []

docs/pages/Guides/Customising-prompts.mdx

@@ -1,49 +1,453 @@
 ---
-title: Customizing Prompts 
-description: This guide will explain how to change prompts in DocsGPT and why it might be benefitial. Additionaly this article expains additional variables that can be used in prompts.
+title: Customizing Prompts
+description: This guide explains how to customize prompts in DocsGPT using the new template-based system with dynamic variable injection.
 ---
 
 import Image from 'next/image'
 
-# Customizing the Main Prompt
+# Customizing Prompts in DocsGPT
 
-Customizing the main prompt for DocsGPT gives you the ability to tailor the AI's responses to your specific requirements. By modifying the prompt text, you can achieve more accurate and relevant answers. Here's how you can do it:
+Customizing prompts for DocsGPT gives you powerful control over the AI's behavior and responses. With the new template-based system, you can inject dynamic context through organized namespaces, making prompts flexible and maintainable without hardcoding values.
+
+## Quick Start
 
 1. Navigate to `SideBar -> Settings`.
-
-
-
-
-2.In Settings select the `Active Prompt` now you will be able to see various prompts style.x
-
-
-
-
-3.Click on the `edit icon`  on the prompt of your choice and you will be able to see the current prompt for it,you can now customise  the prompt as per your choice.
+2. In Settings, select the `Active Prompt` to see various prompt styles.
+3. Click on the `edit icon` on your chosen prompt to customize it.
 
 ### Video Demo
 <Image src="/prompts.gif" alt="prompts" width={800} height={500} />
 
+---
 
+## Template-Based Prompt System
 
-## Example Prompt Modification
+DocsGPT now uses **Jinja2 templating** with four organized namespaces for dynamic variable injection:
+
+### Available Namespaces
+
+#### 1. **`system`** - System Metadata
+Access system-level information:
+
+```jinja
+{{ system.date }}         # Current date (YYYY-MM-DD)
+{{ system.time }}         # Current time (HH:MM:SS)
+{{ system.timestamp }}    # ISO 8601 timestamp
+{{ system.request_id }}   # Unique request identifier
+{{ system.user_id }}      # Current user ID
+```
+
+#### 2. **`source`** - Retrieved Documents
+Access RAG (Retrieval-Augmented Generation) document context:
+
+```jinja
+{{ source.content }}      # Concatenated document content
+{{ source.summaries }}    # Alias for content (backward compatible)
+{{ source.documents }}    # List of document objects
+{{ source.count }}        # Number of retrieved documents
+```
+
+#### 3. **`passthrough`** - Request Parameters
+Access custom parameters passed in the API request:
+
+```jinja
+{{ passthrough.company }}     # Custom field from request
+{{ passthrough.user_name }}   # User-provided data
+{{ passthrough.context }}     # Any custom parameter
+```
+
+To use passthrough data, send it in your API request:
+```json
+{
+  "question": "What is the pricing?",
+  "passthrough": {
+    "company": "Acme Corp",
+    "user_name": "Alice",
+    "plan_type": "enterprise"
+  }
+}
+```
+
+#### 4. **`tools`** - Pre-fetched Tool Data
+Access results from tools that run before the agent (like memory tool):
+
+```jinja
+{{ tools.memory.root }}       # Memory tool directory listing
+{{ tools.memory.available }}  # Boolean: is memory available
+```
+
+---
+
+## Example Prompts
+
+### Basic Prompt with Documents
+```jinja
+You are a helpful AI assistant for DocsGPT.
+
+Current date: {{ system.date }}
+
+Use the following documents to answer the question:
+
+{{ source.content }}
+
+Provide accurate, helpful answers with code examples when relevant.
+```
+
+### Advanced Prompt with All Namespaces
+```jinja
+You are an AI assistant for {{ passthrough.company }}.
+
+**System Info:**
+- Date: {{ system.date }}
+- Request ID: {{ system.request_id }}
+
+**User Context:**
+- User: {{ passthrough.user_name }}
+- Role: {{ passthrough.role }}
+
+**Available Documents ({{ source.count }}):**
+{{ source.content }}
+
+**Memory Context:**
+{% if tools.memory.available %}
+{{ tools.memory.root }}
+{% else %}
+No saved context available.
+{% endif %}
+
+Please provide detailed, accurate answers based on the documents above.
+```
+
+### Conditional Logic Example
+```jinja
+You are a DocsGPT assistant.
+
+{% if source.count > 0 %}
+I found {{ source.count }} relevant document(s):
+
+{{ source.content }}
+
+Base your answer on these documents.
+{% else %}
+No documents were found. Please answer based on your general knowledge.
+{% endif %}
+```
+
+---
+
+## Migration Guide
+
+### Legacy Format (Still Supported)
+The old `{summaries}` format continues to work for backward compatibility:
 
-**Original Prompt:**
 ```markdown
-You are a DocsGPT, friendly and helpful AI assistant by Arc53 that provides help with documents. You give thorough answers with code examples if possible.
-Use the following pieces of context to help answer the users question. If it's not relevant to the question, provide friendly responses.
-You have access to chat history, and can use it to help answer the question.
-When using code examples, use the following format:
+You are a helpful assistant.
 
-(code)
+Documents:
 {summaries}
 ```
 
-Note that `{summaries}` allows model to see and respond to your upploaded documents. If you don't want this functionality you can safely remove it from the customized prompt.
+This will automatically substitute `{summaries}` with document content.
 
-Feel free to customize the prompt to align it with your specific use case or the kind of responses you want from the AI. For example, you can focus on specific document types, industries, or topics to get more targeted results.
+### New Template Format (Recommended)
+Migrate to the new template syntax for more flexibility:
+
+```jinja
+You are a helpful assistant.
+
+Documents:
+{{ source.content }}
+```
+
+**Migration mapping:**
+- `{summaries}` → `{{ source.content }}` or `{{ source.summaries }}`
+
+---
+
+## Best Practices
+
+### 1. **Use Descriptive Context**
+```jinja
+**Retrieved Documents:**
+{{ source.content }}
+
+**User Query Context:**
+- Company: {{ passthrough.company }}
+- Department: {{ passthrough.department }}
+```
+
+### 2. **Handle Missing Data Gracefully**
+```jinja
+{% if passthrough.user_name %}
+Hello {{ passthrough.user_name }}!
+{% endif %}
+```
+
+### 3. **Leverage Memory for Continuity**
+```jinja
+{% if tools.memory.available %}
+**Previous Context:**
+{{ tools.memory.root }}
+{% endif %}
+
+**Current Question:**
+Please consider the above context when answering.
+```
+
+### 4. **Add Clear Instructions**
+```jinja
+You are a technical support assistant.
+
+**Guidelines:**
+1. Always reference the documents below
+2. Provide step-by-step instructions
+3. Include code examples when relevant
+
+**Reference Documents:**
+{{ source.content }}
+```
+
+---
+
+## Advanced Features
+
+### Looping Over Documents
+```jinja
+{% for doc in source.documents %}
+**Source {{ loop.index }}:** {{ doc.filename }}
+{{ doc.text }}
+
+{% endfor %}
+```
+
+### Date-Based Behavior
+```jinja
+{% if system.date > "2025-01-01" %}
+Note: This is information from 2025 or later.
+{% endif %}
+```
+
+### Custom Formatting
+```jinja
+**Request Information**
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+• Request ID: {{ system.request_id }}
+• User: {{ passthrough.user_name | default("Guest") }}
+• Time: {{ system.time }}
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+---
+
+## Tool Pre-Fetching
+
+### Memory Tool Configuration
+Enable memory tool pre-fetching to inject saved context into prompts:
+
+```python
+# In your tool configuration
+{
+  "name": "memory",
+  "config": {
+    "pre_fetch_enabled": true  # Default: true
+  }
+}
+```
+
+Control pre-fetching globally:
+```bash
+# .env file
+ENABLE_TOOL_PREFETCH=true
+```
+
+Or per-request:
+```json
+{
+  "question": "What are the requirements?",
+  "disable_tool_prefetch": false
+}
+```
+
+---
+
+## Debugging Prompts
+
+### View Rendered Prompts in Logs
+Set log level to `INFO` to see the final rendered prompt sent to the LLM:
+
+```bash
+export LOG_LEVEL=INFO
+```
+
+You'll see output like:
+```
+INFO - Rendered system prompt for agent (length: 1234 chars):
+================================================================================
+You are a helpful assistant for Acme Corp.
+
+Current date: 2025-10-30
+Request ID: req_abc123
+
+Documents:
+Technical documentation about...
+================================================================================
+```
+
+### Template Validation
+Test your template syntax before saving:
+```python
+from application.api.answer.services.prompt_renderer import PromptRenderer
+
+renderer = PromptRenderer()
+is_valid = renderer.validate_template("Your prompt with {{ variables }}")
+```
+
+---
+
+## Common Use Cases
+
+### 1. Customer Support Bot
+```jinja
+You are a customer support assistant for {{ passthrough.company }}.
+
+**Customer:** {{ passthrough.customer_name }}
+**Ticket ID:** {{ system.request_id }}
+**Date:** {{ system.date }}
+
+**Knowledge Base:**
+{{ source.content }}
+
+**Previous Interactions:**
+{{ tools.memory.root }}
+
+Please provide helpful, friendly support based on the knowledge base above.
+```
+
+### 2. Technical Documentation Assistant
+```jinja
+You are a technical documentation expert.
+
+**Available Documentation ({{ source.count }} documents):**
+{{ source.content }}
+
+**Requirements:**
+- Provide code examples in {{ passthrough.language }}
+- Focus on {{ passthrough.framework }} best practices
+- Include relevant links when possible
+```
+
+### 3. Internal Knowledge Base
+```jinja
+You are an internal AI assistant for {{ passthrough.department }}.
+
+**Employee:** {{ passthrough.employee_name }}
+**Access Level:** {{ passthrough.access_level }}
+
+**Relevant Documents:**
+{{ source.content }}
+
+Provide detailed answers appropriate for {{ passthrough.access_level }} access level.
+```
+
+---
+
+## Template Syntax Reference
+
+### Variables
+```jinja
+{{ variable_name }}              # Output variable
+{{ namespace.field }}            # Access nested field
+{{ variable | default("N/A") }}  # Default value
+```
+
+### Conditionals
+```jinja
+{% if condition %}
+  Content
+{% elif other_condition %}
+  Other content
+{% else %}
+  Default content
+{% endif %}
+```
+
+### Loops
+```jinja
+{% for item in list %}
+  {{ item.field }}
+{% endfor %}
+```
+
+### Comments
+```jinja
+{# This is a comment and won't appear in output #}
+```
+
+---
+
+## Security Considerations
+
+1. **Input Sanitization**: Passthrough data is automatically sanitized to prevent injection attacks
+2. **Type Filtering**: Only primitive types (string, int, float, bool, None) are allowed in passthrough
+3. **Autoescaping**: Jinja2 autoescaping is enabled by default
+4. **Size Limits**: Consider the token budget when including large documents
+
+---
+
+## Troubleshooting
+
+### Problem: Variables Not Rendering
+**Solution:** Ensure you're using the correct namespace:
+```jinja
+❌ {{ company }}
+✅ {{ passthrough.company }}
+```
+
+### Problem: Empty Output for Tool Data
+**Solution:** Check that tool pre-fetching is enabled and the tool is configured correctly.
+
+### Problem: Syntax Errors
+**Solution:** Validate template syntax. Common issues:
+```jinja
+❌ {{ variable }     # Missing closing brace
+❌ {% if x %         # Missing closing %}
+✅ {{ variable }}
+✅ {% if x %}...{% endif %}
+```
+
+### Problem: Legacy Prompts Not Working
+**Solution:** The system auto-detects template syntax. If your prompt uses `{summaries}`, it will work in legacy mode. To use new features, add `{{ }}` syntax.
+
+---
+
+## API Reference
+
+### Render Prompt via API
+```python
+from application.api.answer.services.prompt_renderer import PromptRenderer
+
+renderer = PromptRenderer()
+rendered = renderer.render_prompt(
+    prompt_content="Your template with {{ passthrough.name }}",
+    user_id="user_123",
+    request_id="req_456",
+    passthrough_data={"name": "Alice"},
+    docs_together="Document content here",
+    tools_data={"memory": {"root": "Files: notes.txt"}}
+)
+```
+
+---
 
 ## Conclusion
 
-Customizing the main prompt for DocsGPT allows you to tailor the AI's responses to your unique requirements. Whether you need in-depth explanations, code examples, or specific insights, you can achieve it by modifying the main prompt. Remember to experiment and fine-tune your prompts to get the best results.
+The new template-based prompt system provides powerful flexibility while maintaining backward compatibility. By leveraging namespaces, you can create dynamic, context-aware prompts that adapt to your specific use case.
 
+**Key Benefits:**
+- ✅ Dynamic variable injection
+- ✅ Organized namespaces
+- ✅ Backward compatible
+- ✅ Security built-in
+- ✅ Easy to debug
+
+Start with simple templates and gradually add complexity as needed. Happy prompting! 🚀

docs/pages/quickstart.mdx

@@ -57,7 +57,7 @@ The easiest way to launch DocsGPT is using the provided `setup.sh` script. This
 
    * **4) Connect Cloud API Provider:** This option lets you connect DocsGPT to a commercial Cloud API provider such as OpenAI, Google (Vertex AI/Gemini), Anthropic (Claude), Groq, HuggingFace Inference API, or Azure OpenAI. You will need an API key from your chosen provider. Select this if you prefer to use a powerful cloud-based LLM.
 
-   * **5) Modify DocsGPT's source code and rebuild the Docker images locally. Instead of pulling prebuilt images from Docker Hub or using the hosted/public API, you build the entire backend and frontend from source, customizing how DocsGPT works internally, or run it in an environment without internet access.
+   * **5) Modify DocsGPT's source code and rebuild the Docker images locally.** Instead of pulling prebuilt images from Docker Hub or using the hosted/public API, you build the entire backend and frontend from source, customizing how DocsGPT works internally, or run it in an environment without internet access.
 
    After selecting an option and providing any required information (like API keys or model names), the script will configure your `.env` file and start DocsGPT using Docker Compose.
 
@@ -119,4 +119,4 @@ If you prefer a more manual approach, you can follow our [Docker Deployment docu
 
 For more advanced customization of DocsGPT settings, such as configuring vector stores, embedding models, and other parameters, please refer to the [DocsGPT Settings documentation](/Deploying/DocsGPT-Settings). This guide explains how to modify the `.env` file or `settings.py` for deeper configuration.
 
-Enjoy using DocsGPT!
+Enjoy using DocsGPT!

frontend/.env.development

@@ -3,4 +3,4 @@ VITE_BASE_URL=http://localhost:5173
 VITE_API_HOST=http://127.0.0.1:7091
 VITE_API_STREAMING=true
 VITE_NOTIFICATION_TEXT="What's new in 0.14.0 — Changelog"
-VITE_NOTIFICATION_LINK="#"
+VITE_NOTIFICATION_LINK="https://blog.docsgpt.cloud/docsgpt-0-14-agents-automate-integrate-and-innovate/"

frontend/.eslintignore

@@ -1,17 +0,0 @@
-node_modules/
-dist/
-prettier.config.cjs
-.eslintrc.cjs
-env.d.ts
-public/
-assets/
-vite-env.d.ts
-.prettierignore
-package-lock.json
-package.json
-postcss.config.cjs
-prettier.config.cjs
-tailwind.config.cjs
-tsconfig.json
-tsconfig.node.json
-vite.config.ts

frontend/.eslintrc.cjs

@@ -1,45 +0,0 @@
-module.exports = {
-  env: {
-    browser: true,
-    es2021: true,
-    node: true,
-  },
-  extends: [
-    'eslint:recommended',
-    'plugin:@typescript-eslint/recommended',
-    'plugin:react/recommended',
-    'plugin:prettier/recommended',
-  ],
-  overrides: [],
-  parser: '@typescript-eslint/parser',
-  parserOptions: {
-    ecmaVersion: 'latest',
-    sourceType: 'module',
-  },
-  plugins: ['react', 'unused-imports'],
-  rules: {
-    'react/prop-types': 'off',
-    'unused-imports/no-unused-imports': 'error',
-    'react/react-in-jsx-scope': 'off',
-    'prettier/prettier': [
-      'error',
-      {
-        endOfLine: 'auto',
-      },
-    ],
-  },
-  settings: {
-    'import/parsers': {
-      '@typescript-eslint/parser': ['.ts', '.tsx'],
-    },
-    react: {
-      version: 'detect',
-    },
-    'import/resolver': {
-      node: {
-        paths: ['src'],
-        extensions: ['.js', '.jsx', '.ts', '.tsx'],
-      },
-    },
-  },
-};

frontend/eslint.config.js

@@ -0,0 +1,78 @@
+import js from '@eslint/js'
+import tsParser from '@typescript-eslint/parser'
+import tsPlugin from '@typescript-eslint/eslint-plugin'
+import react from 'eslint-plugin-react'
+import unusedImports from 'eslint-plugin-unused-imports'
+import prettier from 'eslint-plugin-prettier'
+import globals from 'globals'
+
+export default [
+  {
+    ignores: [
+      'node_modules/',
+      'dist/',
+      'prettier.config.cjs',
+      '.eslintrc.cjs',
+      'env.d.ts',
+      'public/',
+      'assets/',
+      'vite-env.d.ts',
+      '.prettierignore',
+      'package-lock.json',
+      'package.json',
+      'postcss.config.cjs',
+      'tailwind.config.cjs',
+      'tsconfig.json',
+      'tsconfig.node.json',
+      'vite.config.ts',
+    ],
+  },
+  {
+    files: ['**/*.{js,jsx,ts,tsx}'],
+    languageOptions: {
+      ecmaVersion: 'latest',
+      sourceType: 'module',
+      parser: tsParser,
+      parserOptions: {
+        ecmaFeatures: {
+          jsx: true,
+        },
+      },
+      globals: {
+        ...globals.browser,
+        ...globals.es2021,
+        ...globals.node,
+      },
+    },
+    plugins: {
+      '@typescript-eslint': tsPlugin,
+      react,
+      'unused-imports': unusedImports,
+      prettier,
+    },
+    rules: {
+      ...js.configs.recommended.rules,
+      ...tsPlugin.configs.recommended.rules,
+      ...react.configs.recommended.rules,
+      ...prettier.configs.recommended.rules,
+      'react/prop-types': 'off',
+      'unused-imports/no-unused-imports': 'error',
+      'react/react-in-jsx-scope': 'off',
+      'no-undef': 'off',
+      '@typescript-eslint/no-explicit-any': 'warn',
+      '@typescript-eslint/no-unused-vars': 'warn',
+      '@typescript-eslint/no-unused-expressions': 'warn',
+      'prettier/prettier': [
+        'error',
+        {
+          endOfLine: 'auto',
+        },
+      ],
+    },
+    settings: {
+      react: {
+        version: 'detect',
+      },
+    },
+  },
+]

frontend/package.json

@@ -19,21 +19,21 @@
     ]
   },
   "dependencies": {
-    "@reduxjs/toolkit": "^2.8.2",
+    "@reduxjs/toolkit": "^2.10.1",
     "chart.js": "^4.4.4",
     "clsx": "^2.1.1",
     "copy-to-clipboard": "^3.3.3",
     "i18next": "^25.5.3",
-    "i18next-browser-languagedetector": "^8.0.2",
+    "i18next-browser-languagedetector": "^8.2.0",
     "lodash": "^4.17.21",
-    "mermaid": "^11.12.0",
+    "mermaid": "^11.12.1",
     "prop-types": "^15.8.1",
     "react": "^19.1.0",
     "react-chartjs-2": "^5.3.0",
-    "react-dom": "^19.0.0",
+    "react-dom": "^19.1.1",
     "react-dropzone": "^14.3.8",
     "react-google-drive-picker": "^1.2.2",
-    "react-i18next": "^15.4.0",
+    "react-i18next": "^16.2.4",
     "react-markdown": "^9.0.1",
     "react-redux": "^9.2.0",
     "react-router-dom": "^7.6.1",
@@ -46,30 +46,28 @@
   "devDependencies": {
     "@tailwindcss/postcss": "^4.1.10",
     "@types/lodash": "^4.17.20",
-    "@types/mermaid": "^9.1.0",
     "@types/react": "^19.1.8",
-    "@types/react-dom": "^19.0.0",
+    "@types/react-dom": "^19.1.7",
     "@types/react-syntax-highlighter": "^15.5.13",
-    "@typescript-eslint/eslint-plugin": "^5.51.0",
-    "@typescript-eslint/parser": "^5.62.0",
+    "@typescript-eslint/eslint-plugin": "^8.46.3",
+    "@typescript-eslint/parser": "^8.46.3",
     "@vitejs/plugin-react": "^4.3.4",
-    "eslint": "^8.57.1",
+    "eslint": "^9.39.1",
     "eslint-config-prettier": "^10.1.5",
-    "eslint-config-standard-with-typescript": "^34.0.0",
     "eslint-plugin-import": "^2.31.0",
-    "eslint-plugin-n": "^15.7.0",
-    "eslint-plugin-prettier": "^5.2.1",
+    "eslint-plugin-n": "^17.23.1",
+    "eslint-plugin-prettier": "^5.5.4",
     "eslint-plugin-promise": "^6.6.0",
     "eslint-plugin-react": "^7.37.5",
     "eslint-plugin-unused-imports": "^4.1.4",
-    "husky": "^8.0.0",
+    "husky": "^9.1.7",
     "lint-staged": "^15.3.0",
     "postcss": "^8.4.49",
     "prettier": "^3.5.3",
-    "prettier-plugin-tailwindcss": "^0.6.13",
+    "prettier-plugin-tailwindcss": "^0.7.1",
     "tailwindcss": "^4.1.11",
     "typescript": "^5.8.3",
-    "vite": "^6.3.5",
+    "vite": "^7.2.0",
     "vite-plugin-svgr": "^4.3.0"
   }
 }

frontend/src/Hero.tsx

@@ -1,6 +1,8 @@
-import DocsGPT3 from './assets/cute_docsgpt3.svg';
 import { useTranslation } from 'react-i18next';
 
+import DocsGPT3 from './assets/cute_docsgpt3.svg';
+import DropdownModel from './components/DropdownModel';
+
 export default function Hero({
   handleQuestion,
 }: {
@@ -26,6 +28,10 @@ export default function Hero({
           <span className="text-4xl font-semibold">DocsGPT</span>
           <img className="mb-1 inline w-14" src={DocsGPT3} alt="docsgpt" />
         </div>
+        {/* Model Selector */}
+        <div className="relative w-72">
+          <DropdownModel />
+        </div>
       </div>
 
       {/* Demo Buttons Section */}
@@ -38,7 +44,7 @@ export default function Hero({
                 <button
                   key={key}
                   onClick={() => handleQuestion({ question: demo.query })}
-                  className={`border-dark-gray text-just-black hover:bg-cultured dark:border-dim-gray dark:text-chinese-white dark:hover:bg-charleston-green w-full rounded-[66px] border bg-transparent px-6 py-[14px] text-left transition-colors ${key >= 2 ? 'hidden md:block' : ''} // Show only 2 buttons on mobile`}
+                  className={`border-dark-gray text-just-black hover:bg-cultured dark:border-dim-gray dark:text-chinese-white dark:hover:bg-charleston-green w-full rounded-[66px] border bg-transparent px-6 py-[14px] text-left transition-colors ${key >= 2 ? 'hidden md:block' : ''}`}
                 >
                   <p className="text-black-1000 dark:text-bright-gray mb-2 font-semibold">
                     {demo.header}

frontend/src/Navigation.tsx

@@ -411,7 +411,9 @@ export default function Navigation({ navOpen, setNavOpen }: NavigationProps) {
           {recentAgents?.length > 0 ? (
             <div>
               <div className="mx-4 my-auto mt-2 flex h-6 items-center">
-                <p className="mt-1 ml-4 text-sm font-semibold">Agents</p>
+                <p className="mt-1 ml-4 text-sm font-semibold">
+                  {t('navigation.agents')}
+                </p>
               </div>
               <div className="agents-container">
                 <div>
@@ -565,7 +567,7 @@ export default function Navigation({ navOpen, setNavOpen }: NavigationProps) {
               <div className="flex items-center gap-1 pr-4">
                 <NavLink
                   target="_blank"
-                  to={'https://discord.gg/WHJdfbQDR4'}
+                  to={'https://discord.gg/vN7YFfdMpj'}
                   className={
                     'rounded-full hover:bg-gray-100 dark:hover:bg-[#28292E]'
                   }

frontend/src/PageNotFound.tsx

@@ -1,13 +1,16 @@
 import { Link } from 'react-router-dom';
+import { useTranslation } from 'react-i18next';
 
 export default function PageNotFound() {
+  const { t } = useTranslation();
+
   return (
     <div className="dark:bg-raisin-black grid min-h-screen">
       <p className="text-jet dark:bg-outer-space mx-auto my-auto mt-20 flex w-full max-w-6xl flex-col place-items-center gap-6 rounded-3xl bg-gray-100 p-6 lg:p-10 xl:p-16 dark:text-gray-100">
-        <h1>404</h1>
-        <p>The page you are looking for does not exist.</p>
+        <h1>{t('pageNotFound.title')}</h1>
+        <p>{t('pageNotFound.message')}</p>
         <button className="pointer-cursor bg-blue-1000 hover:bg-blue-3000 mr-4 flex cursor-pointer items-center justify-center rounded-full px-4 py-2 text-white transition-colors duration-100">
-          <Link to="/">Go Back Home</Link>
+          <Link to="/">{t('pageNotFound.goHome')}</Link>
         </button>
       </p>
     </div>

frontend/src/agents/AgentLogs.tsx

@@ -1,4 +1,5 @@
 import { useEffect, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import { useSelector } from 'react-redux';
 import { useNavigate, useParams } from 'react-router-dom';
 
@@ -11,6 +12,7 @@ import Logs from '../settings/Logs';
 import { Agent } from './types';
 
 export default function AgentLogs() {
+  const { t } = useTranslation();
   const navigate = useNavigate();
   const { agentId } = useParams();
   const token = useSelector(selectToken);
@@ -45,12 +47,12 @@ export default function AgentLogs() {
           <img src={ArrowLeft} alt="left-arrow" className="h-3 w-3" />
         </button>
         <p className="text-eerie-black dark:text-bright-gray mt-px text-sm font-semibold">
-          Back to all agents
+          {t('agents.backToAll')}
         </p>
       </div>
       <div className="mt-5 flex w-full flex-wrap items-center justify-between gap-2 px-4">
         <h1 className="text-eerie-black m-0 text-[32px] font-bold md:text-[40px] dark:text-white">
-          Agent Logs
+          {t('agents.logs.title')}
         </h1>
       </div>
       <div className="mt-6 flex flex-col gap-3 px-4">
@@ -59,9 +61,10 @@ export default function AgentLogs() {
             <p className="text-[#28292E] dark:text-[#E0E0E0]">{agent.name}</p>
             <p className="text-xs text-[#28292E] dark:text-[#E0E0E0]/40">
               {agent.last_used_at
-                ? 'Last used at ' +
+                ? t('agents.logs.lastUsedAt') +
+                  ' ' +
                   new Date(agent.last_used_at).toLocaleString()
-                : 'No usage history'}
+                : t('agents.logs.noUsageHistory')}
             </p>
           </div>
         )}
@@ -79,7 +82,9 @@ export default function AgentLogs() {
           <Spinner />
         </div>
       ) : (
-        agent && <Logs agentId={agent.id} tableHeader="Agent endpoint logs" />
+        agent && (
+          <Logs agentId={agent.id} tableHeader={t('agents.logs.tableHeader')} />
+        )
       )}
     </div>
   );

frontend/src/agents/AgentPreview.tsx

@@ -1,4 +1,5 @@
 import { useCallback, useEffect, useRef, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import { useDispatch, useSelector } from 'react-redux';
 
 import MessageInput from '../components/MessageInput';
@@ -17,6 +18,7 @@ import { selectSelectedAgent } from '../preferences/preferenceSlice';
 import { AppDispatch } from '../store';
 
 export default function AgentPreview() {
+  const { t } = useTranslation();
   const dispatch = useDispatch<AppDispatch>();
 
   const queries = useSelector(selectPreviewQueries);
@@ -130,8 +132,7 @@ export default function AgentPreview() {
           />
         </div>
         <p className="text-gray-4000 dark:text-sonic-silver w-full bg-transparent text-center text-xs md:inline">
-          This is a preview of the agent. You can publish it to start using it
-          in conversations.
+          {t('agents.preview.testMessage')}
         </p>
       </div>
     </div>

frontend/src/agents/AgentsList.tsx

@@ -1,4 +1,5 @@
 import { useEffect, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import { useDispatch, useSelector } from 'react-redux';
 import { useNavigate } from 'react-router-dom';
 
@@ -17,6 +18,7 @@ import { agentSectionsConfig } from './agents.config';
 import { Agent } from './types';
 
 export default function AgentsList() {
+  const { t } = useTranslation();
   const dispatch = useDispatch();
   const token = useSelector(selectToken);
   const selectedAgent = useSelector(selectSelectedAgent);
@@ -33,11 +35,10 @@ export default function AgentsList() {
   return (
     <div className="p-4 md:p-12">
       <h1 className="text-eerie-black mb-0 text-[32px] font-bold lg:text-[40px] dark:text-[#E0E0E0]">
-        Agents
+        {t('agents.title')}
       </h1>
       <p className="dark:text-gray-4000 mt-5 text-[15px] text-[#71717A]">
-        Discover and create custom versions of DocsGPT that combine
-        instructions, extra knowledge, and any combination of skills
+        {t('agents.description')}
       </p>
       {agentSectionsConfig.map((sectionConfig) => (
         <AgentSection key={sectionConfig.id} config={sectionConfig} />
@@ -51,6 +52,7 @@ function AgentSection({
 }: {
   config: (typeof agentSectionsConfig)[number];
 }) {
+  const { t } = useTranslation();
   const navigate = useNavigate();
   const dispatch = useDispatch();
   const token = useSelector(selectToken);
@@ -85,16 +87,18 @@ function AgentSection({
       <div className="flex w-full items-center justify-between">
         <div className="flex flex-col gap-2">
           <h2 className="text-[18px] font-semibold text-[#18181B] dark:text-[#E0E0E0]">
-            {config.title}
+            {t(`agents.sections.${config.id}.title`)}
           </h2>
-          <p className="text-[13px] text-[#71717A]">{config.description}</p>
+          <p className="text-[13px] text-[#71717A]">
+            {t(`agents.sections.${config.id}.description`)}
+          </p>
         </div>
         {config.showNewAgentButton && (
           <button
             className="bg-purple-30 hover:bg-violets-are-blue rounded-full px-4 py-2 text-sm text-white"
             onClick={() => navigate('/agents/new')}
           >
-            New Agent
+            {t('agents.newAgent')}
           </button>
         )}
       </div>
@@ -117,13 +121,13 @@ function AgentSection({
           </div>
         ) : (
           <div className="flex h-72 w-full flex-col items-center justify-center gap-3 text-base text-[#18181B] dark:text-[#E0E0E0]">
-            <p>{config.emptyStateDescription}</p>
+            <p>{t(`agents.sections.${config.id}.emptyState`)}</p>
             {config.showNewAgentButton && (
               <button
                 className="bg-purple-30 hover:bg-violets-are-blue ml-2 rounded-full px-4 py-2 text-sm text-white"
                 onClick={() => navigate('/agents/new')}
               >
-                New Agent
+                {t('agents.newAgent')}
               </button>
             )}
           </div>

frontend/src/agents/NewAgent.tsx

@@ -1,8 +1,10 @@
 import isEqual from 'lodash/isEqual';
 import React, { useCallback, useEffect, useRef, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import { useDispatch, useSelector } from 'react-redux';
 import { useNavigate, useParams } from 'react-router-dom';
 
+import modelService from '../api/services/modelService';
 import userService from '../api/services/userService';
 import ArrowLeft from '../assets/arrow-left.svg';
 import SourceIcon from '../assets/source.svg';
@@ -25,11 +27,13 @@ import { UserToolType } from '../settings/types';
 import AgentPreview from './AgentPreview';
 import { Agent, ToolSummary } from './types';
 
+import type { Model } from '../models/types';
 const embeddingsName =
   import.meta.env.VITE_EMBEDDINGS_NAME ||
   'huggingface_sentence-transformers/all-mpnet-base-v2';
 
 export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
+  const { t } = useTranslation();
   const navigate = useNavigate();
   const dispatch = useDispatch();
   const { agentId } = useParams();
@@ -57,18 +61,25 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
     token_limit: undefined,
     limited_request_mode: false,
     request_limit: undefined,
+    models: [],
+    default_model_id: '',
   });
   const [imageFile, setImageFile] = useState<File | null>(null);
   const [prompts, setPrompts] = useState<
     { name: string; id: string; type: string }[]
   >([]);
   const [userTools, setUserTools] = useState<OptionType[]>([]);
+  const [availableModels, setAvailableModels] = useState<Model[]>([]);
   const [isSourcePopupOpen, setIsSourcePopupOpen] = useState(false);
   const [isToolsPopupOpen, setIsToolsPopupOpen] = useState(false);
+  const [isModelsPopupOpen, setIsModelsPopupOpen] = useState(false);
   const [selectedSourceIds, setSelectedSourceIds] = useState<
     Set<string | number>
   >(new Set());
   const [selectedTools, setSelectedTools] = useState<ToolSummary[]>([]);
+  const [selectedModelIds, setSelectedModelIds] = useState<Set<string>>(
+    new Set(),
+  );
   const [deleteConfirmation, setDeleteConfirmation] =
     useState<ActiveState>('INACTIVE');
   const [agentDetails, setAgentDetails] = useState<ActiveState>('INACTIVE');
@@ -84,11 +95,12 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
   const initialAgentRef = useRef<Agent | null>(null);
   const sourceAnchorButtonRef = useRef<HTMLButtonElement>(null);
   const toolAnchorButtonRef = useRef<HTMLButtonElement>(null);
+  const modelAnchorButtonRef = useRef<HTMLButtonElement>(null);
 
   const modeConfig = {
     new: {
-      heading: 'New Agent',
-      buttonText: 'Publish',
+      heading: t('agents.form.headings.new'),
+      buttonText: t('agents.form.buttons.publish'),
       showDelete: false,
       showSaveDraft: true,
       showLogs: false,
@@ -96,8 +108,8 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       trackChanges: false,
     },
     edit: {
-      heading: 'Edit Agent',
-      buttonText: 'Save',
+      heading: t('agents.form.headings.edit'),
+      buttonText: t('agents.form.buttons.save'),
       showDelete: true,
       showSaveDraft: false,
       showLogs: true,
@@ -105,8 +117,8 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       trackChanges: true,
     },
     draft: {
-      heading: 'New Agent (Draft)',
-      buttonText: 'Publish',
+      heading: t('agents.form.headings.draft'),
+      buttonText: t('agents.form.buttons.publish'),
       showDelete: true,
       showSaveDraft: true,
       showLogs: false,
@@ -116,8 +128,8 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
   };
   const chunks = ['0', '2', '4', '6', '8', '10'];
   const agentTypes = [
-    { label: 'Classic', value: 'classic' },
-    { label: 'ReAct', value: 'react' },
+    { label: t('agents.form.agentTypes.classic'), value: 'classic' },
+    { label: t('agents.form.agentTypes.react'), value: 'react' },
   ];
 
   const isPublishable = () => {
@@ -198,13 +210,19 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
 
     if (agent.limited_token_mode && agent.token_limit) {
       formData.append('limited_token_mode', 'True');
-      formData.append('token_limit', JSON.stringify(agent.token_limit));
-    } else formData.append('token_limit', '0');
+      formData.append('token_limit', agent.token_limit.toString());
+    } else {
+      formData.append('limited_token_mode', 'False');
+      formData.append('token_limit', '0');
+    }
 
     if (agent.limited_request_mode && agent.request_limit) {
       formData.append('limited_request_mode', 'True');
-      formData.append('request_limit', JSON.stringify(agent.request_limit));
-    } else formData.append('request_limit', '0');
+      formData.append('request_limit', agent.request_limit.toString());
+    } else {
+      formData.append('limited_request_mode', 'False');
+      formData.append('request_limit', '0');
+    }
 
     if (imageFile) formData.append('image', imageFile);
 
@@ -216,6 +234,13 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       formData.append('json_schema', JSON.stringify(agent.json_schema));
     }
 
+    if (agent.models && agent.models.length > 0) {
+      formData.append('models', JSON.stringify(agent.models));
+    }
+    if (agent.default_model_id) {
+      formData.append('default_model_id', agent.default_model_id);
+    }
+
     try {
       setDraftLoading(true);
       const response =
@@ -295,15 +320,29 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       formData.append('json_schema', JSON.stringify(agent.json_schema));
     }
 
+    // Always send the limited mode fields
     if (agent.limited_token_mode && agent.token_limit) {
       formData.append('limited_token_mode', 'True');
-      formData.append('token_limit', JSON.stringify(agent.token_limit));
-    } else formData.append('token_limit', '0');
+      formData.append('token_limit', agent.token_limit.toString());
+    } else {
+      formData.append('limited_token_mode', 'False');
+      formData.append('token_limit', '0');
+    }
 
     if (agent.limited_request_mode && agent.request_limit) {
       formData.append('limited_request_mode', 'True');
-      formData.append('request_limit', JSON.stringify(agent.request_limit));
-    } else formData.append('request_limit', '0');
+      formData.append('request_limit', agent.request_limit.toString());
+    } else {
+      formData.append('limited_request_mode', 'False');
+      formData.append('request_limit', '0');
+    }
+
+    if (agent.models && agent.models.length > 0) {
+      formData.append('models', JSON.stringify(agent.models));
+    }
+    if (agent.default_model_id) {
+      formData.append('default_model_id', agent.default_model_id);
+    }
 
     try {
       setPublishLoading(true);
@@ -373,8 +412,16 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       const data = await response.json();
       setPrompts(data);
     };
+    const getModels = async () => {
+      const response = await modelService.getModels(null);
+      if (!response.ok) throw new Error('Failed to fetch models');
+      const data = await response.json();
+      const transformed = modelService.transformModels(data.models || []);
+      setAvailableModels(transformed);
+    };
     getTools();
     getPrompts();
+    getModels();
   }, [token]);
 
   // Auto-select default source if none selected
@@ -447,6 +494,34 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
     }
   }, [agentId, mode, token]);
 
+  useEffect(() => {
+    if (agent.models && agent.models.length > 0 && availableModels.length > 0) {
+      const agentModelIds = new Set(agent.models);
+      if (agentModelIds.size > 0 && selectedModelIds.size === 0) {
+        setSelectedModelIds(agentModelIds);
+      }
+    }
+  }, [agent.models, availableModels.length]);
+
+  useEffect(() => {
+    const modelsArray = Array.from(selectedModelIds);
+    if (modelsArray.length > 0) {
+      setAgent((prev) => ({
+        ...prev,
+        models: modelsArray,
+        default_model_id: modelsArray.includes(prev.default_model_id || '')
+          ? prev.default_model_id
+          : modelsArray[0],
+      }));
+    } else {
+      setAgent((prev) => ({
+        ...prev,
+        models: [],
+        default_model_id: '',
+      }));
+    }
+  }, [selectedModelIds]);
+
   useEffect(() => {
     const selectedSources = Array.from(selectedSourceIds)
       .map((id) =>
@@ -543,7 +618,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
           <img src={ArrowLeft} alt="left-arrow" className="h-3 w-3" />
         </button>
         <p className="text-eerie-black dark:text-bright-gray mt-px text-sm font-semibold">
-          Back to all agents
+          {t('agents.backToAll')}
         </p>
       </div>
       <div className="mt-5 flex w-full flex-wrap items-center justify-between gap-2 px-4">
@@ -555,7 +630,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
             className="text-purple-30 dark:text-light-gray mr-4 rounded-3xl py-2 text-sm font-medium dark:bg-transparent"
             onClick={handleCancel}
           >
-            Cancel
+            {t('agents.form.buttons.cancel')}
           </button>
           {modeConfig[effectiveMode].showDelete && agent.id && (
             <button
@@ -563,7 +638,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
               onClick={() => setDeleteConfirmation('ACTIVE')}
             >
               <span className="block h-4 w-4 bg-[url('/src/assets/red-trash.svg')] bg-contain bg-center bg-no-repeat transition-all group-hover:bg-[url('/src/assets/white-trash.svg')]" />
-              Delete
+              {t('agents.form.buttons.delete')}
             </button>
           )}
           {modeConfig[effectiveMode].showSaveDraft && (
@@ -578,7 +653,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                 {draftLoading ? (
                   <Spinner size="small" color="#976af3" />
                 ) : (
-                  'Save Draft'
+                  t('agents.form.buttons.saveDraft')
                 )}
               </span>
             </button>
@@ -589,7 +664,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
               onClick={() => navigate(`/agents/logs/${agent.id}`)}
             >
               <span className="block h-5 w-5 bg-[url('/src/assets/monitoring-purple.svg')] bg-contain bg-center bg-no-repeat transition-all group-hover:bg-[url('/src/assets/monitoring-white.svg')]" />
-              Logs
+              {t('agents.form.buttons.logs')}
             </button>
           )}
           {modeConfig[effectiveMode].showAccessDetails && (
@@ -597,7 +672,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
               className="hover:bg-vi</button>olets-are-blue border-violets-are-blue text-violets-are-blue hover:bg-violets-are-blue rounded-3xl border border-solid px-5 py-2 text-sm font-medium transition-colors hover:text-white"
               onClick={() => setAgentDetails('ACTIVE')}
             >
-              Access Details
+              {t('agents.form.buttons.accessDetails')}
             </button>
           )}
           <button
@@ -618,17 +693,19 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
       <div className="mt-3 flex w-full flex-1 grid-cols-5 flex-col gap-10 rounded-[30px] bg-[#F6F6F6] p-5 max-[1179px]:overflow-visible min-[1180px]:grid min-[1180px]:gap-5 min-[1180px]:overflow-hidden dark:bg-[#383838]">
         <div className="scrollbar-thin col-span-2 flex flex-col gap-5 max-[1179px]:overflow-visible min-[1180px]:max-h-full min-[1180px]:overflow-y-auto min-[1180px]:pr-3">
           <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
-            <h2 className="text-lg font-semibold">Meta</h2>
+            <h2 className="text-lg font-semibold">
+              {t('agents.form.sections.meta')}
+            </h2>
             <input
               className="border-silver text-jet dark:bg-raisin-black dark:text-bright-gray dark:placeholder:text-silver mt-3 w-full rounded-3xl border bg-white px-5 py-3 text-sm outline-hidden placeholder:text-gray-400 dark:border-[#7E7E7E]"
               type="text"
               value={agent.name}
-              placeholder="Agent name"
+              placeholder={t('agents.form.placeholders.agentName')}
               onChange={(e) => setAgent({ ...agent, name: e.target.value })}
             />
             <textarea
               className="border-silver text-jet dark:bg-raisin-black dark:text-bright-gray dark:placeholder:text-silver mt-3 h-32 w-full rounded-xl border bg-white px-5 py-4 text-sm outline-hidden placeholder:text-gray-400 dark:border-[#7E7E7E]"
-              placeholder="Describe your agent"
+              placeholder={t('agents.form.placeholders.describeAgent')}
               value={agent.description}
               onChange={(e) =>
                 setAgent({ ...agent, description: e.target.value })
@@ -641,9 +718,12 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                 onUpload={handleUpload}
                 onRemove={() => setImageFile(null)}
                 uploadText={[
-                  { text: 'Click to upload', colorClass: 'text-[#7D54D1]' },
                   {
-                    text: ' or drag and drop',
+                    text: t('agents.form.upload.clickToUpload'),
+                    colorClass: 'text-[#7D54D1]',
+                  },
+                  {
+                    text: t('agents.form.upload.dragAndDrop'),
                     colorClass: 'text-[#525252]',
                   },
                 ]}
@@ -651,7 +731,9 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
             </div>
           </div>
           <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
-            <h2 className="text-lg font-semibold">Source</h2>
+            <h2 className="text-lg font-semibold">
+              {t('agents.form.sections.source')}
+            </h2>
             <div className="mt-3">
               <div className="flex flex-wrap items-center gap-1">
                 <button
@@ -672,11 +754,13 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                               source.name === id ||
                               source.retriever === id,
                           );
-                          return matchedDoc?.name || `External KB`;
+                          return (
+                            matchedDoc?.name || t('agents.form.externalKb')
+                          );
                         })
                         .filter(Boolean)
                         .join(', ')
-                    : 'Select sources'}
+                    : t('agents.form.placeholders.selectSources')}
                 </button>
                 <MultiSelectPopup
                   isOpen={isSourcePopupOpen}
@@ -720,9 +804,13 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       setSelectedSourceIds(newSelectedIds);
                     }
                   }}
-                  title="Select Sources"
-                  searchPlaceholder="Search sources..."
-                  noOptionsMessage="No sources available"
+                  title={t('agents.form.sourcePopup.title')}
+                  searchPlaceholder={t(
+                    'agents.form.sourcePopup.searchPlaceholder',
+                  )}
+                  noOptionsMessage={t(
+                    'agents.form.sourcePopup.noOptionsMessage',
+                  )}
                 />
               </div>
               <div className="mt-3">
@@ -737,7 +825,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                   border="border"
                   buttonClassName="bg-white dark:bg-[#222327] border-silver dark:border-[#7E7E7E]"
                   optionsClassName="bg-white dark:bg-[#383838] border-silver dark:border-[#7E7E7E]"
-                  placeholder="Chunks per query"
+                  placeholder={t('agents.form.placeholders.chunksPerQuery')}
                   placeholderClassName="text-gray-400 dark:text-silver"
                   contentSize="text-sm"
                 />
@@ -757,7 +845,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                     setAgent({ ...agent, prompt_id: id })
                   }
                   setPrompts={setPrompts}
-                  title="Prompt"
+                  title={t('agents.form.sections.prompt')}
                   titleClassName="text-lg font-semibold dark:text-[#E0E0E0]"
                   showAddButton={false}
                   dropdownProps={{
@@ -777,12 +865,14 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                 className="border-violets-are-blue text-violets-are-blue hover:bg-violets-are-blue w-20 shrink-0 basis-full rounded-3xl border-2 border-solid px-5 py-[11px] text-sm transition-colors hover:text-white sm:basis-auto"
                 onClick={() => setAddPromptModal('ACTIVE')}
               >
-                Add
+                {t('agents.form.buttons.add')}
               </button>
             </div>
           </div>
           <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
-            <h2 className="text-lg font-semibold">Tools</h2>
+            <h2 className="text-lg font-semibold">
+              {t('agents.form.sections.tools')}
+            </h2>
             <div className="mt-3 flex flex-wrap items-center gap-1">
               <button
                 ref={toolAnchorButtonRef}
@@ -798,7 +888,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       .map((tool) => tool.display_name || tool.name)
                       .filter(Boolean)
                       .join(', ')
-                  : 'Select tools'}
+                  : t('agents.form.placeholders.selectTools')}
               </button>
               <MultiSelectPopup
                 isOpen={isToolsPopupOpen}
@@ -817,14 +907,18 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       })),
                   )
                 }
-                title="Select Tools"
-                searchPlaceholder="Search tools..."
-                noOptionsMessage="No tools available"
+                title={t('agents.form.toolsPopup.title')}
+                searchPlaceholder={t(
+                  'agents.form.toolsPopup.searchPlaceholder',
+                )}
+                noOptionsMessage={t('agents.form.toolsPopup.noOptionsMessage')}
               />
             </div>
           </div>
           <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
-            <h2 className="text-lg font-semibold">Agent type</h2>
+            <h2 className="text-lg font-semibold">
+              {t('agents.form.sections.agentType')}
+            </h2>
             <div className="mt-3">
               <Dropdown
                 options={agentTypes}
@@ -842,12 +936,88 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                 border="border"
                 buttonClassName="bg-white dark:bg-[#222327] border-silver dark:border-[#7E7E7E]"
                 optionsClassName="bg-white dark:bg-[#383838] border-silver dark:border-[#7E7E7E]"
-                placeholder="Select type"
+                placeholder={t('agents.form.placeholders.selectType')}
                 placeholderClassName="text-gray-400 dark:text-silver"
                 contentSize="text-sm"
               />
             </div>
           </div>
+          <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
+            <h2 className="text-lg font-semibold">
+              {t('agents.form.sections.models')}
+            </h2>
+            <div className="mt-3 flex flex-col gap-3">
+              <button
+                ref={modelAnchorButtonRef}
+                onClick={() => setIsModelsPopupOpen(!isModelsPopupOpen)}
+                className={`border-silver dark:bg-raisin-black w-full truncate rounded-3xl border bg-white px-5 py-3 text-left text-sm dark:border-[#7E7E7E] ${
+                  selectedModelIds.size > 0
+                    ? 'text-jet dark:text-bright-gray'
+                    : 'dark:text-silver text-gray-400'
+                }`}
+              >
+                {selectedModelIds.size > 0
+                  ? availableModels
+                      .filter((m) => selectedModelIds.has(m.id))
+                      .map((m) => m.display_name)
+                      .join(', ')
+                  : t('agents.form.placeholders.selectModels')}
+              </button>
+              <MultiSelectPopup
+                isOpen={isModelsPopupOpen}
+                onClose={() => setIsModelsPopupOpen(false)}
+                anchorRef={modelAnchorButtonRef}
+                options={availableModels.map((model) => ({
+                  id: model.id,
+                  label: model.display_name,
+                }))}
+                selectedIds={selectedModelIds}
+                onSelectionChange={(newSelectedIds: Set<string | number>) =>
+                  setSelectedModelIds(
+                    new Set(Array.from(newSelectedIds).map(String)),
+                  )
+                }
+                title={t('agents.form.modelsPopup.title')}
+                searchPlaceholder={t(
+                  'agents.form.modelsPopup.searchPlaceholder',
+                )}
+                noOptionsMessage={t('agents.form.modelsPopup.noOptionsMessage')}
+              />
+              {selectedModelIds.size > 0 && (
+                <div>
+                  <label className="mb-2 block text-sm font-medium">
+                    {t('agents.form.labels.defaultModel')}
+                  </label>
+                  <Dropdown
+                    options={availableModels
+                      .filter((m) => selectedModelIds.has(m.id))
+                      .map((m) => ({
+                        label: m.display_name,
+                        value: m.id,
+                      }))}
+                    selectedValue={
+                      availableModels.find(
+                        (m) => m.id === agent.default_model_id,
+                      )?.display_name || null
+                    }
+                    onSelect={(option: { label: string; value: string }) =>
+                      setAgent({ ...agent, default_model_id: option.value })
+                    }
+                    size="w-full"
+                    rounded="3xl"
+                    border="border"
+                    buttonClassName="bg-white dark:bg-[#222327] border-silver dark:border-[#7E7E7E]"
+                    optionsClassName="bg-white dark:bg-[#383838] border-silver dark:border-[#7E7E7E]"
+                    placeholder={t(
+                      'agents.form.placeholders.selectDefaultModel',
+                    )}
+                    placeholderClassName="text-gray-400 dark:text-silver"
+                    contentSize="text-sm"
+                  />
+                </div>
+              )}
+            </div>
+          </div>
           <div className="dark:bg-raisin-black rounded-[30px] bg-white px-6 py-3 dark:text-[#E0E0E0]">
             <button
               onClick={() =>
@@ -856,7 +1026,9 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
               className="flex w-full items-center justify-between text-left focus:outline-none"
             >
               <div>
-                <h2 className="text-lg font-semibold">Advanced</h2>
+                <h2 className="text-lg font-semibold">
+                  {t('agents.form.sections.advanced')}
+                </h2>
               </div>
               <div className="ml-4 flex items-center">
                 <svg
@@ -879,9 +1051,11 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
             {isAdvancedSectionExpanded && (
               <div className="mt-3">
                 <div>
-                  <h2 className="text-sm font-medium">JSON response schema</h2>
+                  <h2 className="text-sm font-medium">
+                    {t('agents.form.advanced.jsonSchema')}
+                  </h2>
                   <p className="mt-1 text-xs text-gray-600 dark:text-gray-400">
-                    Define a JSON schema to enforce structured output format
+                    {t('agents.form.advanced.jsonSchemaDescription')}
                   </p>
                 </div>
                 <textarea
@@ -915,17 +1089,19 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       }`}
                     />
                     {jsonSchemaValid
-                      ? 'Valid JSON'
-                      : 'Invalid JSON - fix to enable saving'}
+                      ? t('agents.form.advanced.validJson')
+                      : t('agents.form.advanced.invalidJson')}
                   </div>
                 )}
 
                 <div className="mt-6">
                   <div className="flex items-center justify-between">
                     <div>
-                      <h2 className="text-sm font-medium">Token limiting</h2>
+                      <h2 className="text-sm font-medium">
+                        {t('agents.form.advanced.tokenLimiting')}
+                      </h2>
                       <p className="mt-1 text-xs text-gray-600 dark:text-gray-400">
-                        Limit daily total tokens that can be used by this agent
+                        {t('agents.form.advanced.tokenLimitingDescription')}
                       </p>
                     </div>
                     <button
@@ -965,7 +1141,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       })
                     }
                     disabled={!agent.limited_token_mode}
-                    placeholder="Enter token limit"
+                    placeholder={t('agents.form.placeholders.enterTokenLimit')}
                     className={`border-silver text-jet dark:bg-raisin-black dark:text-bright-gray dark:placeholder:text-silver mt-2 w-full rounded-3xl border bg-white px-5 py-3 text-sm outline-hidden placeholder:text-gray-400 dark:border-[#7E7E7E] ${
                       !agent.limited_token_mode
                         ? 'cursor-not-allowed opacity-50'
@@ -977,10 +1153,11 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                 <div className="mt-6">
                   <div className="flex items-center justify-between">
                     <div>
-                      <h2 className="text-sm font-medium">Request limiting</h2>
+                      <h2 className="text-sm font-medium">
+                        {t('agents.form.advanced.requestLimiting')}
+                      </h2>
                       <p className="mt-1 text-xs text-gray-600 dark:text-gray-400">
-                        Limit daily total requests that can be made to this
-                        agent
+                        {t('agents.form.advanced.requestLimitingDescription')}
                       </p>
                     </div>
                     <button
@@ -1020,7 +1197,9 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
                       })
                     }
                     disabled={!agent.limited_request_mode}
-                    placeholder="Enter request limit"
+                    placeholder={t(
+                      'agents.form.placeholders.enterRequestLimit',
+                    )}
                     className={`border-silver text-jet dark:bg-raisin-black dark:text-bright-gray dark:placeholder:text-silver mt-2 w-full rounded-3xl border bg-white px-5 py-3 text-sm outline-hidden placeholder:text-gray-400 dark:border-[#7E7E7E] ${
                       !agent.limited_request_mode
                         ? 'cursor-not-allowed opacity-50'
@@ -1033,22 +1212,24 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
           </div>
         </div>
         <div className="col-span-3 flex flex-col gap-2 max-[1179px]:h-auto max-[1179px]:px-0 max-[1179px]:py-0 min-[1180px]:h-full min-[1180px]:py-2 dark:text-[#E0E0E0]">
-          <h2 className="text-lg font-semibold">Preview</h2>
+          <h2 className="text-lg font-semibold">
+            {t('agents.form.sections.preview')}
+          </h2>
           <div className="flex-1 max-[1179px]:overflow-visible min-[1180px]:min-h-0 min-[1180px]:overflow-hidden">
             <AgentPreviewArea />
           </div>
         </div>
       </div>
       <ConfirmationModal
-        message="Are you sure you want to delete this agent?"
+        message={t('agents.deleteConfirmation')}
         modalState={deleteConfirmation}
         setModalState={setDeleteConfirmation}
-        submitLabel="Delete"
+        submitLabel={t('agents.form.buttons.delete')}
         handleSubmit={() => {
           handleDelete(agent.id || '');
           setDeleteConfirmation('INACTIVE');
         }}
-        cancelLabel="Cancel"
+        cancelLabel={t('agents.form.buttons.cancel')}
         variant="danger"
       />
       <AgentDetailsModal
@@ -1071,6 +1252,7 @@ export default function NewAgent({ mode }: { mode: 'new' | 'edit' | 'draft' }) {
 }
 
 function AgentPreviewArea() {
+  const { t } = useTranslation();
   const selectedAgent = useSelector(selectSelectedAgent);
   return (
     <div className="dark:bg-raisin-black w-full rounded-[30px] border border-[#F6F6F6] bg-white max-[1179px]:h-[600px] min-[1180px]:h-full dark:border-[#7E7E7E]">
@@ -1082,7 +1264,7 @@ function AgentPreviewArea() {
         <div className="flex h-full w-full flex-col items-center justify-center gap-2">
           <span className="block h-12 w-12 bg-[url('/src/assets/science-spark.svg')] bg-contain bg-center bg-no-repeat transition-all dark:bg-[url('/src/assets/science-spark-dark.svg')]" />{' '}
           <p className="dark:text-gray-4000 text-xs text-[#18181B]">
-            Published agents can be previewed here
+            {t('agents.form.preview.publishedPreview')}
           </p>
         </div>
       )}

frontend/src/agents/SharedAgent.tsx

@@ -144,7 +144,7 @@ export default function SharedAgent() {
             className="mx-auto mb-6 h-32 w-32"
           />
           <p className="dark:text-gray-4000 text-center text-lg text-[#71717A]">
-            No agent found. Please ensure the agent is shared.
+            {t('agents.shared.notFound')}
           </p>
         </div>
       </div>

frontend/src/agents/agentPreviewSlice.ts

@@ -52,6 +52,10 @@ export const fetchPreviewAnswer = createAsyncThunk<
     }
 
     if (state.preference) {
+      const modelId =
+        state.preference.selectedAgent?.default_model_id ||
+        state.preference.selectedModel?.id;
+
       if (API_STREAMING) {
         await handleFetchAnswerSteaming(
           question,
@@ -120,22 +124,23 @@ export const fetchPreviewAnswer = createAsyncThunk<
           indx,
           state.preference.selectedAgent?.id,
           attachmentIds,
-          false, // Don't save preview conversations
+          false,
+          modelId,
         );
       } else {
-        // Non-streaming implementation
         const answer = await handleFetchAnswer(
           question,
           signal,
           state.preference.token,
           state.preference.selectedDocs,
-          null, // No conversation ID for previews
+          null,
           state.preference.prompt.id,
           state.preference.chunks,
           state.preference.token_limit,
           state.preference.selectedAgent?.id,
           attachmentIds,
-          false, // Don't save preview conversations
+          false,
+          modelId,
         );
 
         if (answer) {

frontend/src/agents/types/index.ts

@@ -32,4 +32,6 @@ export type Agent = {
   token_limit?: number;
   limited_request_mode?: boolean;
   request_limit?: number;
+  models?: string[];
+  default_model_id?: string;
 };

frontend/src/api/endpoints.ts

@@ -2,6 +2,7 @@ const endpoints = {
   USER: {
     CONFIG: '/api/config',
     NEW_TOKEN: '/api/generate_token',
+    MODELS: '/api/models',
     DOCS: '/api/sources',
     DOCS_PAGINATED: '/api/sources/paginated',
     API_KEYS: '/api/get_api_keys',

frontend/src/api/services/modelService.ts

@@ -0,0 +1,25 @@
+import apiClient from '../client';
+import endpoints from '../endpoints';
+
+import type { AvailableModel, Model } from '../../models/types';
+
+const modelService = {
+  getModels: (token: string | null): Promise<Response> =>
+    apiClient.get(endpoints.USER.MODELS, token, {}),
+
+  transformModels: (models: AvailableModel[]): Model[] =>
+    models.map((model) => ({
+      id: model.id,
+      value: model.id,
+      provider: model.provider,
+      display_name: model.display_name,
+      description: model.description,
+      context_window: model.context_window,
+      supported_attachment_types: model.supported_attachment_types,
+      supports_tools: model.supports_tools,
+      supports_structured_output: model.supports_structured_output,
+      supports_streaming: model.supports_streaming,
+    })),
+};
+
+export default modelService;

frontend/src/assets/book.svg

@@ -0,0 +1,3 @@
+<svg width="12" height="14" viewBox="0 0 12 14" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path d="M10.2857 14H2.57143C1.15179 14 0 12.8242 0 11.375V2.625C0 1.17578 1.15179 0 2.57143 0H10.7143C11.4241 0 12 0.587891 12 1.3125V9.1875C12 9.75898 11.6411 10.2457 11.1429 10.4262V12.25C11.617 12.25 12 12.641 12 13.125C12 13.609 11.617 14 11.1429 14H10.2857ZM2.57143 10.5C2.09732 10.5 1.71429 10.891 1.71429 11.375C1.71429 11.859 2.09732 12.25 2.57143 12.25H9.42857V10.5H2.57143ZM3.42857 4.15625C3.42857 4.51992 3.71518 4.8125 4.07143 4.8125H8.78571C9.14196 4.8125 9.42857 4.51992 9.42857 4.15625C9.42857 3.79258 9.14196 3.5 8.78571 3.5H4.07143C3.71518 3.5 3.42857 3.79258 3.42857 4.15625ZM4.07143 6.125C3.71518 6.125 3.42857 6.41758 3.42857 6.78125C3.42857 7.14492 3.71518 7.4375 4.07143 7.4375H8.78571C9.14196 7.4375 9.42857 7.14492 9.42857 6.78125C9.42857 6.41758 9.14196 6.125 8.78571 6.125H4.07143Z" fill="#6A4DF4"/>
+</svg>

frontend/src/assets/rounded-tick.svg

@@ -0,0 +1,3 @@
+<svg width="20" height="21" viewBox="0 0 20 21" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path d="M10 0.75C4.62391 0.75 0.25 5.12391 0.25 10.5C0.25 15.8761 4.62391 20.25 10 20.25C15.3761 20.25 19.75 15.8761 19.75 10.5C19.75 5.12391 15.3761 0.75 10 0.75ZM15.0742 7.23234L8.77422 14.7323C8.70511 14.8147 8.61912 14.8812 8.52207 14.9273C8.42502 14.9735 8.31918 14.9983 8.21172 15H8.19906C8.09394 15 7.99 14.9778 7.89398 14.935C7.79797 14.8922 7.71202 14.8297 7.64172 14.7516L4.94172 11.7516C4.87315 11.6788 4.81981 11.5931 4.78483 11.4995C4.74986 11.4059 4.73395 11.3062 4.73805 11.2063C4.74215 11.1064 4.76617 11.0084 4.8087 10.9179C4.85124 10.8275 4.91142 10.7464 4.98572 10.6796C5.06002 10.6127 5.14694 10.5614 5.24136 10.5286C5.33579 10.4958 5.43581 10.4822 5.53556 10.4886C5.63531 10.495 5.73277 10.5213 5.82222 10.5659C5.91166 10.6106 5.99128 10.6726 6.05641 10.7484L8.17938 13.1072L13.9258 6.26766C14.0547 6.11863 14.237 6.02631 14.4335 6.01066C14.6299 5.99501 14.8246 6.05728 14.9754 6.18402C15.1263 6.31075 15.2212 6.49176 15.2397 6.68793C15.2582 6.8841 15.1988 7.07966 15.0742 7.23234Z" fill="#B5B5B5"/>
+</svg>

frontend/src/components/ActionButtons.tsx

@@ -45,7 +45,7 @@ export default function ActionButtons({
       <div className={`flex items-center gap-2 sm:gap-4 ${className}`}>
         {showNewChat && (
           <button
-            title="Open New Chat"
+            title={t('actionButtons.openNewChat')}
             onClick={newChat}
             className="hover:bg-bright-gray flex items-center gap-1 rounded-full p-2 lg:hidden dark:hover:bg-[#28292E]"
           >
@@ -62,7 +62,7 @@ export default function ActionButtons({
         {showShare && conversationId && (
           <>
             <button
-              title="Share"
+              title={t('actionButtons.share')}
               onClick={() => setShareModalState(true)}
               className="hover:bg-bright-gray rounded-full p-2 dark:hover:bg-[#28292E]"
             >

frontend/src/components/ConnectorTree.tsx

@@ -38,7 +38,7 @@ interface DirectoryStructure {
   [key: string]: FileNode;
 }
 
-interface ConnectorTreeComponentProps {
+interface ConnectorTreeProps {
   docId: string;
   sourceName: string;
   onBackToDocuments: () => void;
@@ -50,7 +50,7 @@ interface SearchResult {
   isFile: boolean;
 }
 
-const ConnectorTreeComponent: React.FC<ConnectorTreeComponentProps> = ({
+const ConnectorTree: React.FC<ConnectorTreeProps> = ({
   docId,
   sourceName,
   onBackToDocuments,
@@ -744,4 +744,4 @@ const ConnectorTreeComponent: React.FC<ConnectorTreeComponentProps> = ({
   );
 };
 
-export default ConnectorTreeComponent;
+export default ConnectorTree;

frontend/src/components/Dropdown.tsx

@@ -60,7 +60,7 @@ function Dropdown<T extends DropdownOption>({
         }`}
       >
         {typeof selectedValue === 'string' ? (
-          <span className="dark:text-bright-gray truncate">
+          <span className={`dark:text-bright-gray truncate ${contentSize}`}>
             {selectedValue}
           </span>
         ) : (

frontend/src/components/DropdownModel.tsx

@@ -0,0 +1,138 @@
+import React, { useEffect } from 'react';
+import { useDispatch, useSelector } from 'react-redux';
+
+import modelService from '../api/services/modelService';
+import Arrow2 from '../assets/dropdown-arrow.svg';
+import RoundedTick from '../assets/rounded-tick.svg';
+import {
+  selectAvailableModels,
+  selectSelectedModel,
+  setAvailableModels,
+  setModelsLoading,
+  setSelectedModel,
+} from '../preferences/preferenceSlice';
+
+import type { Model } from '../models/types';
+
+export default function DropdownModel() {
+  const dispatch = useDispatch();
+  const selectedModel = useSelector(selectSelectedModel);
+  const availableModels = useSelector(selectAvailableModels);
+  const dropdownRef = React.useRef<HTMLDivElement>(null);
+  const [isOpen, setIsOpen] = React.useState(false);
+
+  useEffect(() => {
+    const loadModels = async () => {
+      if ((availableModels?.length ?? 0) > 0) {
+        return;
+      }
+      dispatch(setModelsLoading(true));
+      try {
+        const response = await modelService.getModels(null);
+        if (!response.ok) {
+          throw new Error(`API error: ${response.status}`);
+        }
+        const data = await response.json();
+        const models = data.models || [];
+        const transformed = modelService.transformModels(models);
+
+        dispatch(setAvailableModels(transformed));
+        if (!selectedModel && transformed.length > 0) {
+          const defaultModel =
+            transformed.find((m) => m.id === data.default_model_id) ||
+            transformed[0];
+          dispatch(setSelectedModel(defaultModel));
+        } else if (selectedModel && transformed.length > 0) {
+          const isValid = transformed.find((m) => m.id === selectedModel.id);
+          if (!isValid) {
+            const defaultModel =
+              transformed.find((m) => m.id === data.default_model_id) ||
+              transformed[0];
+            dispatch(setSelectedModel(defaultModel));
+          }
+        }
+      } catch (error) {
+        console.error('Failed to load models:', error);
+      } finally {
+        dispatch(setModelsLoading(false));
+      }
+    };
+
+    loadModels();
+  }, [availableModels?.length, dispatch, selectedModel]);
+
+  const handleClickOutside = (event: MouseEvent) => {
+    if (
+      dropdownRef.current &&
+      !dropdownRef.current.contains(event.target as Node)
+    ) {
+      setIsOpen(false);
+    }
+  };
+
+  useEffect(() => {
+    document.addEventListener('mousedown', handleClickOutside);
+    return () => {
+      document.removeEventListener('mousedown', handleClickOutside);
+    };
+  }, []);
+
+  return (
+    <div ref={dropdownRef}>
+      <div
+        className={`bg-gray-1000 dark:bg-dark-charcoal mx-auto flex w-full cursor-pointer justify-between p-1 dark:text-white ${isOpen ? 'rounded-t-3xl' : 'rounded-3xl'}`}
+        onClick={() => setIsOpen(!isOpen)}
+      >
+        {selectedModel?.display_name ? (
+          <p className="mx-4 my-3 truncate overflow-hidden whitespace-nowrap">
+            {selectedModel.display_name}
+          </p>
+        ) : (
+          <p className="mx-4 my-3 truncate overflow-hidden whitespace-nowrap">
+            Select Model
+          </p>
+        )}
+        <img
+          src={Arrow2}
+          alt="arrow"
+          className={`${
+            isOpen ? 'rotate-360' : 'rotate-270'
+          } mr-3 w-3 transition-all select-none`}
+        />
+      </div>
+      {isOpen && (
+        <div className="no-scrollbar dark:bg-dark-charcoal absolute right-0 left-0 z-20 -mt-1 max-h-52 w-full overflow-y-auto rounded-b-3xl bg-white shadow-md">
+          {availableModels && (availableModels?.length ?? 0) > 0 ? (
+            availableModels.map((model: Model) => (
+              <div
+                key={model.id}
+                onClick={() => {
+                  dispatch(setSelectedModel(model));
+                  setIsOpen(false);
+                }}
+                className={`border-gray-3000/75 dark:border-purple-taupe/50 hover:bg-gray-3000/75 dark:hover:bg-purple-taupe flex h-10 w-full cursor-pointer items-center justify-between border-t`}
+              >
+                <div className="flex w-full items-center justify-between">
+                  <p className="overflow-hidden py-3 pr-2 pl-5 overflow-ellipsis whitespace-nowrap">
+                    {model.display_name}
+                  </p>
+                  {model.id === selectedModel?.id ? (
+                    <img
+                      src={RoundedTick}
+                      alt="selected"
+                      className="mr-3.5 h-4 w-4"
+                    />
+                  ) : null}
+                </div>
+              </div>
+            ))
+          ) : (
+            <div className="h-10 w-full border-x-2 border-b-2">
+              <p className="ml-5 py-3 text-gray-500">No models available</p>
+            </div>
+          )}
+        </div>
+      )}
+    </div>
+  );
+}

frontend/src/components/FilePicker.tsx

@@ -1,4 +1,5 @@
 import React, { useState, useEffect, useCallback, useRef } from 'react';
+import { useTranslation } from 'react-i18next';
 import { formatBytes } from '../utils/stringUtils';
 import { formatDate } from '../utils/dateTimeUtils';
 import {
@@ -66,6 +67,7 @@ export const FilePicker: React.FC<CloudFilePickerProps> = ({
     );
   };
 
+  const { t } = useTranslation();
   const [files, setFiles] = useState<CloudFile[]>([]);
   const [selectedFiles, setSelectedFiles] =
     useState<string[]>(initialSelectedFiles);
@@ -417,7 +419,7 @@ export const FilePicker: React.FC<CloudFilePickerProps> = ({
               <div className="mb-3 max-w-md">
                 <Input
                   type="text"
-                  placeholder="Search files and folders..."
+                  placeholder={t('filePicker.searchPlaceholder')}
                   value={searchQuery}
                   onChange={(e) => handleSearchChange(e.target.value)}
                   colorVariant="silver"
@@ -431,7 +433,9 @@ export const FilePicker: React.FC<CloudFilePickerProps> = ({
 
               {/* Selected Files Message */}
               <div className="pb-3 text-sm text-gray-600 dark:text-gray-400">
-                {selectedFiles.length + selectedFolders.length} selected
+                {t('filePicker.itemsSelected', {
+                  count: selectedFiles.length + selectedFolders.length,
+                })}
               </div>
             </div>
 
@@ -448,9 +452,15 @@ export const FilePicker: React.FC<CloudFilePickerProps> = ({
                       <TableHead>
                         <TableRow>
                           <TableHeader width="40px"></TableHeader>
-                          <TableHeader width="60%">Name</TableHeader>
-                          <TableHeader width="20%">Last Modified</TableHeader>
-                          <TableHeader width="20%">Size</TableHeader>
+                          <TableHeader width="60%">
+                            {t('filePicker.name')}
+                          </TableHeader>
+                          <TableHeader width="20%">
+                            {t('filePicker.lastModified')}
+                          </TableHeader>
+                          <TableHeader width="20%">
+                            {t('filePicker.size')}
+                          </TableHeader>
                         </TableRow>
                       </TableHead>
                       <TableBody>

frontend/src/components/FileTree.tsx

@@ -36,7 +36,7 @@ interface DirectoryStructure {
   [key: string]: FileNode;
 }
 
-interface FileTreeComponentProps {
+interface FileTreeProps {
   docId: string;
   sourceName: string;
   onBackToDocuments: () => void;
@@ -48,7 +48,7 @@ interface SearchResult {
   isFile: boolean;
 }
 
-const FileTreeComponent: React.FC<FileTreeComponentProps> = ({
+const FileTree: React.FC<FileTreeProps> = ({
   docId,
   sourceName,
   onBackToDocuments,
@@ -871,4 +871,4 @@ const FileTreeComponent: React.FC<FileTreeComponentProps> = ({
   );
 };
 
-export default FileTreeComponent;
+export default FileTree;

frontend/src/components/FileUpload.tsx

@@ -1,4 +1,5 @@
 import React, { useCallback, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import { useDropzone } from 'react-dropzone';
 import { twMerge } from 'tailwind-merge';
 
@@ -44,13 +45,14 @@ export const FileUpload = ({
   activeClassName = 'border-blue-500 bg-blue-50',
   acceptClassName = 'border-green-500 dark:border-green-500 bg-green-50 dark:bg-green-50/10',
   rejectClassName = 'border-red-500 bg-red-50 dark:bg-red-500/10 dark:border-red-500',
-  uploadText = 'Click to upload or drag and drop',
-  dragActiveText = 'Drop the files here',
-  fileTypeText = 'PNG, JPG, JPEG up to',
-  sizeLimitText = 'MB',
+  uploadText,
+  dragActiveText,
+  fileTypeText,
+  sizeLimitText,
   disabled = false,
   validator,
 }: FileUploadProps) => {
+  const { t } = useTranslation();
   const [errors, setErrors] = useState<string[]>([]);
   const [preview, setPreview] = useState<string | null>(null);
   const [currentFile, setCurrentFile] = useState<File | null>(null);
@@ -71,7 +73,9 @@ export const FileUpload = ({
     if (file.size > maxSize) {
       return {
         isValid: false,
-        error: `File exceeds ${maxSize / 1024 / 1024}MB limit`,
+        error: t('components.fileUpload.fileSizeError', {
+          size: maxSize / 1024 / 1024,
+        }),
       };
     }
 
@@ -178,7 +182,11 @@ export const FileUpload = ({
         </p>
       );
     }
-    return <p className="text-sm font-semibold">{uploadText}</p>;
+    return (
+      <p className="text-sm font-semibold">
+        {uploadText || t('components.fileUpload.clickToUpload')}
+      </p>
+    );
   };
 
   const defaultContent = (
@@ -196,14 +204,17 @@ export const FileUpload = ({
       <div className="text-center">
         <div className="text-sm font-medium">
           {isDragActive ? (
-            <p className="text-sm font-semibold">{dragActiveText}</p>
+            <p className="text-sm font-semibold">
+              {dragActiveText || t('components.fileUpload.dropFiles')}
+            </p>
           ) : (
             renderUploadText()
           )}
         </div>
         <p className="mt-1 text-xs text-[#A3A3A3]">
-          {fileTypeText} {maxSize / 1024 / 1024}
-          {sizeLimitText}
+          {fileTypeText || t('components.fileUpload.fileTypes')}{' '}
+          {maxSize / 1024 / 1024}
+          {sizeLimitText || t('components.fileUpload.sizeLimitUnit')}
         </p>
       </div>
     </div>

frontend/src/components/Input.tsx

@@ -20,6 +20,7 @@ const Input = ({
   onChange,
   onPaste,
   onKeyDown,
+  edgeRoundness = 'rounded-full',
 }: InputProps) => {
   const colorStyles = {
     silver: 'border-silver dark:border-silver/40',
@@ -43,7 +44,7 @@ const Input = ({
     <div className={`relative ${className}`}>
       <input
         ref={inputRef}
-        className={`peer text-jet dark:text-bright-gray h-[42px] w-full rounded-full bg-transparent ${leftIcon ? 'pl-10' : 'px-3'} py-1 placeholder-transparent outline-hidden ${colorStyles[colorVariant]} ${borderStyles[borderVariant]} ${textSizeStyles[textSize]} [&:-webkit-autofill]:appearance-none [&:-webkit-autofill]:bg-transparent [&:-webkit-autofill_selected]:bg-transparent`}
+        className={`peer text-jet dark:text-bright-gray h-[42px] w-full ${edgeRoundness} bg-transparent ${leftIcon ? 'pl-10' : 'px-3'} py-1 placeholder-transparent outline-hidden ${colorStyles[colorVariant]} ${borderStyles[borderVariant]} ${textSizeStyles[textSize]} [&:-webkit-autofill]:appearance-none [&:-webkit-autofill]:bg-transparent [&:-webkit-autofill_selected]:bg-transparent`}
         type={type}
         id={id}
         name={name}

frontend/src/components/MermaidRenderer.tsx

@@ -1,4 +1,5 @@
 import React, { useEffect, useRef, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import mermaid from 'mermaid';
 import CopyButton from './CopyButton';
 import { Prism as SyntaxHighlighter } from 'react-syntax-highlighter';
@@ -15,6 +16,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
   code,
   isLoading,
 }) => {
+  const { t } = useTranslation();
   const [isDarkTheme] = useDarkTheme();
   const diagramId = useRef(
     `mermaid-${Date.now()}-${Math.random().toString(36).substring(2)}`,
@@ -273,7 +275,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
               <button
                 onClick={() => setShowDownloadMenu(!showDownloadMenu)}
                 className="flex h-full items-center rounded-sm bg-gray-100 px-2 py-1 text-xs dark:bg-gray-700"
-                title="Download options"
+                title={t('mermaid.downloadOptions')}
               >
                 Download <span className="ml-1">▼</span>
               </button>
@@ -307,7 +309,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
                   ? 'bg-blue-200 dark:bg-blue-800'
                   : 'bg-gray-100 dark:bg-gray-700'
               }`}
-              title="View Code"
+              title={t('mermaid.viewCode')}
             >
               Code
             </button>
@@ -353,7 +355,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
                       setZoomFactor((prev) => Math.max(1, prev - 0.5))
                     }
                     className="rounded px-1 hover:bg-gray-600"
-                    title="Decrease zoom"
+                    title={t('mermaid.decreaseZoom')}
                   >
                     -
                   </button>
@@ -362,7 +364,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
                     onClick={() => {
                       setZoomFactor(2);
                     }}
-                    title="Reset zoom"
+                    title={t('mermaid.resetZoom')}
                   >
                     {zoomFactor.toFixed(1)}x
                   </span>
@@ -371,7 +373,7 @@ const MermaidRenderer: React.FC<MermaidRendererProps> = ({
                       setZoomFactor((prev) => Math.min(6, prev + 0.5))
                     }
                     className="rounded px-1 hover:bg-gray-600"
-                    title="Increase zoom"
+                    title={t('mermaid.increaseZoom')}
                   >
                     +
                   </button>

frontend/src/components/MessageInput.tsx

@@ -19,8 +19,8 @@ import {
   removeAttachment,
   selectAttachments,
   updateAttachment,
+  reorderAttachments,
 } from '../upload/uploadSlice';
-import { reorderAttachments } from '../upload/uploadSlice';
 
 import { ActiveState } from '../models/misc';
 import {
@@ -77,7 +77,7 @@ export default function MessageInput({
         (browserOS === 'mac' && event.metaKey && event.key === 'k')
       ) {
         event.preventDefault();
-        setIsSourcesPopupOpen(!isSourcesPopupOpen);
+        setIsSourcesPopupOpen((s) => !s);
       }
     };
 
@@ -89,8 +89,198 @@ export default function MessageInput({
 
   const uploadFiles = useCallback(
     (files: File[]) => {
+      if (!files || files.length === 0) return;
+
       const apiHost = import.meta.env.VITE_API_HOST;
 
+      if (files.length > 1) {
+        const formData = new FormData();
+        const indexToUiId: Record<number, string> = {};
+
+        files.forEach((file, i) => {
+          formData.append('file', file);
+          const uiId = crypto.randomUUID();
+          indexToUiId[i] = uiId;
+          dispatch(
+            addAttachment({
+              id: uiId,
+              fileName: file.name,
+              progress: 0,
+              status: 'uploading' as const,
+              taskId: '',
+            }),
+          );
+        });
+
+        const xhr = new XMLHttpRequest();
+
+        xhr.upload.addEventListener('progress', (event) => {
+          if (event.lengthComputable) {
+            const progress = Math.round((event.loaded / event.total) * 100);
+            Object.values(indexToUiId).forEach((uiId) =>
+              dispatch(
+                updateAttachment({
+                  id: uiId,
+                  updates: { progress },
+                }),
+              ),
+            );
+          }
+        });
+
+        xhr.onload = () => {
+          const status = xhr.status;
+          if (status === 200) {
+            try {
+              const response = JSON.parse(xhr.responseText);
+
+              if (Array.isArray(response?.tasks)) {
+                const tasks = response.tasks as Array<{
+                  task_id?: string;
+                  filename?: string;
+                  attachment_id?: string;
+                  path?: string;
+                }>;
+
+                tasks.forEach((t, idx) => {
+                  const uiId = indexToUiId[idx];
+                  if (!uiId) return;
+                  if (t?.task_id) {
+                    dispatch(
+                      updateAttachment({
+                        id: uiId,
+                        updates: {
+                          taskId: t.task_id,
+                          status: 'processing',
+                          progress: 10,
+                        },
+                      }),
+                    );
+                  } else {
+                    dispatch(
+                      updateAttachment({
+                        id: uiId,
+                        updates: { status: 'failed' },
+                      }),
+                    );
+                  }
+                });
+
+                if (tasks.length < files.length) {
+                  for (let i = tasks.length; i < files.length; i++) {
+                    const uiId = indexToUiId[i];
+                    if (uiId) {
+                      dispatch(
+                        updateAttachment({
+                          id: uiId,
+                          updates: { status: 'failed' },
+                        }),
+                      );
+                    }
+                  }
+                }
+              } else if (response?.task_id) {
+                if (files.length === 1) {
+                  const uiId = indexToUiId[0];
+                  if (uiId) {
+                    dispatch(
+                      updateAttachment({
+                        id: uiId,
+                        updates: {
+                          taskId: response.task_id,
+                          status: 'processing',
+                          progress: 10,
+                        },
+                      }),
+                    );
+                  }
+                } else {
+                  console.warn(
+                    'Server returned a single task_id for multiple files. Update backend to return tasks[].',
+                  );
+                  const firstUi = indexToUiId[0];
+                  if (firstUi) {
+                    dispatch(
+                      updateAttachment({
+                        id: firstUi,
+                        updates: {
+                          taskId: response.task_id,
+                          status: 'processing',
+                          progress: 10,
+                        },
+                      }),
+                    );
+                  }
+                  for (let i = 1; i < files.length; i++) {
+                    const uiId = indexToUiId[i];
+                    if (uiId) {
+                      dispatch(
+                        updateAttachment({
+                          id: uiId,
+                          updates: { status: 'failed' },
+                        }),
+                      );
+                    }
+                  }
+                }
+              } else {
+                console.error('Unexpected upload response shape', response);
+                Object.values(indexToUiId).forEach((id) =>
+                  dispatch(
+                    updateAttachment({
+                      id,
+                      updates: { status: 'failed' },
+                    }),
+                  ),
+                );
+              }
+            } catch (err) {
+              console.error(
+                'Failed to parse upload response',
+                err,
+                xhr.responseText,
+              );
+              Object.values(indexToUiId).forEach((id) =>
+                dispatch(
+                  updateAttachment({
+                    id,
+                    updates: { status: 'failed' },
+                  }),
+                ),
+              );
+            }
+          } else {
+            console.error('Upload failed', status, xhr.responseText);
+            Object.values(indexToUiId).forEach((id) =>
+              dispatch(
+                updateAttachment({
+                  id,
+                  updates: { status: 'failed' },
+                }),
+              ),
+            );
+          }
+        };
+
+        xhr.onerror = () => {
+          console.error('Upload network error');
+          Object.values(indexToUiId).forEach((id) =>
+            dispatch(
+              updateAttachment({
+                id,
+                updates: { status: 'failed' },
+              }),
+            ),
+          );
+        };
+
+        xhr.open('POST', `${apiHost}${endpoints.USER.STORE_ATTACHMENT}`);
+        if (token) xhr.setRequestHeader('Authorization', `Bearer ${token}`);
+        xhr.send(formData);
+        return;
+      }
+
+      // Single-file path: upload each file individually (original repo behavior)
       files.forEach((file) => {
         const formData = new FormData();
         formData.append('file', file);
@@ -121,16 +311,54 @@ export default function MessageInput({
 
         xhr.onload = () => {
           if (xhr.status === 200) {
-            const response = JSON.parse(xhr.responseText);
-            if (response.task_id) {
+            try {
+              const response = JSON.parse(xhr.responseText);
+              if (response.task_id) {
+                dispatch(
+                  updateAttachment({
+                    id: uniqueId,
+                    updates: {
+                      taskId: response.task_id,
+                      status: 'processing',
+                      progress: 10,
+                    },
+                  }),
+                );
+              } else {
+                // If backend returned tasks[] for single-file, handle gracefully:
+                if (
+                  Array.isArray(response?.tasks) &&
+                  response.tasks[0]?.task_id
+                ) {
+                  dispatch(
+                    updateAttachment({
+                      id: uniqueId,
+                      updates: {
+                        taskId: response.tasks[0].task_id,
+                        status: 'processing',
+                        progress: 10,
+                      },
+                    }),
+                  );
+                } else {
+                  dispatch(
+                    updateAttachment({
+                      id: uniqueId,
+                      updates: { status: 'failed' },
+                    }),
+                  );
+                }
+              }
+            } catch (err) {
+              console.error(
+                'Failed to parse upload response',
+                err,
+                xhr.responseText,
+              );
               dispatch(
                 updateAttachment({
                   id: uniqueId,
-                  updates: {
-                    taskId: response.task_id,
-                    status: 'processing',
-                    progress: 10,
-                  },
+                  updates: { status: 'failed' },
                 }),
               );
             }
@@ -154,7 +382,7 @@ export default function MessageInput({
         };
 
         xhr.open('POST', `${apiHost}${endpoints.USER.STORE_ATTACHMENT}`);
-        xhr.setRequestHeader('Authorization', `Bearer ${token}`);
+        if (token) xhr.setRequestHeader('Authorization', `Bearer ${token}`);
         xhr.send(formData);
       });
     },
@@ -163,15 +391,13 @@ export default function MessageInput({
 
   const handleFileAttachment = (e: React.ChangeEvent<HTMLInputElement>) => {
     if (!e.target.files || e.target.files.length === 0) return;
-
     const files = Array.from(e.target.files);
     uploadFiles(files);
-
     // clear input so same file can be selected again
     e.target.value = '';
   };
 
-  // Drag and drop handler
+  // Drag & drop via react-dropzone
   const onDrop = useCallback(
     (acceptedFiles: File[]) => {
       uploadFiles(acceptedFiles);
@@ -321,11 +547,8 @@ export default function MessageInput({
     handleAbort();
   };
 
-  // Drag state for reordering
   const [draggingId, setDraggingId] = useState<string | null>(null);
 
-  // no preview object URLs to revoke (preview removed per reviewer request)
-
   const findIndexById = (id: string) =>
     attachments.findIndex((a) => a.id === id);
 
@@ -359,7 +582,9 @@ export default function MessageInput({
 
   return (
     <div {...getRootProps()} className="flex w-full flex-col">
+      {/* react-dropzone input (for drag/drop) */}
       <input {...getInputProps()} />
+
       <div className="border-dark-gray bg-lotion dark:border-grey relative flex w-full flex-col rounded-[23px] border dark:bg-transparent">
         <div className="flex flex-wrap gap-1.5 px-2 py-2 sm:gap-2 sm:px-3">
           {attachments.map((attachment) => {
@@ -374,7 +599,11 @@ export default function MessageInput({
                   attachment.status !== 'completed'
                     ? 'opacity-70'
                     : 'opacity-100'
-                } ${draggingId === attachment.id ? 'ring-dashed opacity-60 ring-2 ring-purple-200' : ''}`}
+                } ${
+                  draggingId === attachment.id
+                    ? 'ring-dashed opacity-60 ring-2 ring-purple-200'
+                    : ''
+                }`}
                 title={attachment.fileName}
               >
                 <div className="bg-purple-30 mr-2 flex h-8 w-8 items-center justify-center rounded-md p-1">

frontend/src/components/Notification.tsx

@@ -1,3 +1,4 @@
+import { useTranslation } from 'react-i18next';
 import close from '../assets/cross.svg';
 import rightArrow from '../assets/arrow-full-right.svg';
 import bg from '../assets/notification-bg.jpg';
@@ -13,13 +14,14 @@ export default function Notification({
   notificationLink,
   handleCloseNotification,
 }: NotificationProps) {
+  const { t } = useTranslation();
   return (
     <a
       className="absolute right-2 bottom-6 z-20 flex w-3/4 items-center justify-center gap-2 rounded-lg bg-cover bg-center bg-no-repeat px-2 py-4 sm:right-4 md:w-2/5 lg:w-1/3 xl:w-1/4 2xl:w-1/5"
       style={{ backgroundImage: `url(${bg})` }}
       href={notificationLink}
       target="_blank"
-      aria-label="Notification"
+      aria-label={t('notification.ariaLabel')}
       rel="noreferrer"
     >
       <p className="text-white-3000 text-xs leading-6 font-semibold xl:text-sm xl:leading-7">
@@ -31,7 +33,7 @@ export default function Notification({
 
       <button
         className="absolute top-2 right-2 z-30 h-4 w-4 hover:opacity-70"
-        aria-label="Close notification"
+        aria-label={t('notification.closeAriaLabel')}
         onClick={(e) => {
           e.stopPropagation();
           e.preventDefault();

frontend/src/components/SettingsBar.tsx

@@ -24,6 +24,7 @@ interface SettingsBarProps {
 }
 
 const SettingsBar = ({ setActiveTab, activeTab }: SettingsBarProps) => {
+  const { t } = useTranslation();
   const [hiddenGradient, setHiddenGradient] =
     useState<HiddenGradientType>('left');
   const containerRef = useRef<null | HTMLDivElement>(null);
@@ -60,7 +61,7 @@ const SettingsBar = ({ setActiveTab, activeTab }: SettingsBarProps) => {
         <button
           onClick={() => scrollTabs(-1)}
           className="flex h-6 w-6 items-center justify-center rounded-full transition-all hover:bg-gray-200 dark:hover:bg-gray-700"
-          aria-label="Scroll tabs left"
+          aria-label={t('settings.scrollTabsLeft')}
         >
           <img src={ArrowLeft} alt="left-arrow" className="h-3" />
         </button>
@@ -69,7 +70,7 @@ const SettingsBar = ({ setActiveTab, activeTab }: SettingsBarProps) => {
         ref={containerRef}
         className="no-scrollbar flex snap-x flex-nowrap overflow-x-auto scroll-smooth md:space-x-4"
         role="tablist"
-        aria-label="Settings tabs"
+        aria-label={t('settings.tabsAriaLabel')}
       >
         {tabs.map((tab, index) => (
           <button
@@ -93,7 +94,7 @@ const SettingsBar = ({ setActiveTab, activeTab }: SettingsBarProps) => {
         <button
           onClick={() => scrollTabs(1)}
           className="flex h-6 w-6 items-center justify-center rounded-full hover:bg-gray-200 dark:hover:bg-gray-700"
-          aria-label="Scroll tabs right"
+          aria-label={t('settings.scrollTabsRight')}
         >
           <img src={ArrowRight} alt="right-arrow" className="h-3" />
         </button>

frontend/src/components/SourcesPopup.tsx

@@ -172,11 +172,7 @@ export default function SourcesPopup({
                                     : doc.date !== option.date,
                                 )
                               : [];
-                          dispatch(
-                            setSelectedDocs(
-                              updatedDocs.length > 0 ? updatedDocs : null,
-                            ),
-                          );
+                          dispatch(setSelectedDocs(updatedDocs));
                           handlePostDocumentSelect(
                             updatedDocs.length > 0 ? updatedDocs : null,
                           );

frontend/src/components/TextToSpeechButton.tsx

@@ -1,94 +1,202 @@
-import { useState, useRef } from 'react';
+import { useState, useRef, useEffect } from 'react';
 import Speaker from '../assets/speaker.svg?react';
 import Stopspeech from '../assets/stopspeech.svg?react';
 import LoadingIcon from '../assets/Loading.svg?react'; // Add a loading icon SVG here
+
 const apiHost = import.meta.env.VITE_API_HOST || 'https://docsapi.arc53.com';
 
-export default function SpeakButton({
-  text,
-  colorLight,
-  colorDark,
-}: {
-  text: string;
-  colorLight?: string;
-  colorDark?: string;
-}) {
+let currentlyPlayingAudio: {
+  audio: HTMLAudioElement;
+  stopCallback: () => void;
+} | null = null;
+
+let currentLoadingRequest: {
+  abortController: AbortController;
+  stopLoadingCallback: () => void;
+} | null = null;
+
+// LRU Cache for audio
+const audioCache = new Map<string, string>();
+const MAX_CACHE_SIZE = 10;
+
+function getCachedAudio(text: string): string | undefined {
+  const cached = audioCache.get(text);
+  if (cached) {
+    audioCache.delete(text);
+    audioCache.set(text, cached);
+  }
+  return cached;
+}
+
+function setCachedAudio(text: string, audioBase64: string) {
+  if (audioCache.has(text)) {
+    audioCache.delete(text);
+  }
+  if (audioCache.size >= MAX_CACHE_SIZE) {
+    const firstKey = audioCache.keys().next().value;
+    if (firstKey !== undefined) {
+      audioCache.delete(firstKey);
+    }
+  }
+
+  audioCache.set(text, audioBase64);
+}
+
+export default function SpeakButton({ text }: { text: string }) {
   const [isSpeaking, setIsSpeaking] = useState(false);
   const [isLoading, setIsLoading] = useState(false);
-  const [isSpeakHovered, setIsSpeakHovered] = useState(false);
   const audioRef = useRef<HTMLAudioElement | null>(null);
+  const abortControllerRef = useRef<AbortController | null>(null);
+
+  useEffect(() => {
+    return () => {
+      // Abort any pending fetch request
+      if (abortControllerRef.current) {
+        abortControllerRef.current.abort();
+        abortControllerRef.current = null;
+      }
+
+      // Stop any playing audio
+      if (audioRef.current) {
+        audioRef.current.pause();
+        if (currentlyPlayingAudio?.audio === audioRef.current) {
+          currentlyPlayingAudio = null;
+        }
+        audioRef.current = null;
+      }
+
+      // Clear global loading request if it's this component's
+      if (currentLoadingRequest) {
+        currentLoadingRequest = null;
+      }
+    };
+  }, []);
 
   const handleSpeakClick = async () => {
     if (isSpeaking) {
-      // Stop audio if it's currently playing
       audioRef.current?.pause();
       audioRef.current = null;
+      currentlyPlayingAudio = null;
       setIsSpeaking(false);
       return;
     }
 
+    // Stop any currently playing audio
+    if (currentlyPlayingAudio) {
+      currentlyPlayingAudio.audio.pause();
+      currentlyPlayingAudio.stopCallback();
+      currentlyPlayingAudio = null;
+    }
+
+    // Abort any pending loading request
+    if (currentLoadingRequest) {
+      currentLoadingRequest.abortController.abort();
+      currentLoadingRequest.stopLoadingCallback();
+      currentLoadingRequest = null;
+    }
+
     try {
-      // Set loading state and initiate TTS request
       setIsLoading(true);
+      const cachedAudio = getCachedAudio(text);
+      let audioBase64: string;
 
-      const response = await fetch(apiHost + '/api/tts', {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({ text }),
-      });
-
-      const data = await response.json();
-
-      if (data.success && data.audio_base64) {
-        // Create and play the audio
-        const audio = new Audio(`data:audio/mp3;base64,${data.audio_base64}`);
-        audioRef.current = audio;
-
-        audio.play().then(() => {
-          setIsSpeaking(true);
-          setIsLoading(false);
-
-          // Reset when audio ends
-          audio.onended = () => {
-            setIsSpeaking(false);
-            audioRef.current = null;
-          };
-        });
-      } else {
-        console.error('Failed to retrieve audio.');
+      if (cachedAudio) {
+        audioBase64 = cachedAudio;
         setIsLoading(false);
+      } else {
+        const abortController = new AbortController();
+        abortControllerRef.current = abortController;
+
+        currentLoadingRequest = {
+          abortController,
+          stopLoadingCallback: () => {
+            setIsLoading(false);
+          },
+        };
+
+        const response = await fetch(apiHost + '/api/tts', {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({ text }),
+          signal: abortController.signal,
+        });
+
+        const data = await response.json();
+        abortControllerRef.current = null;
+        currentLoadingRequest = null;
+
+        if (data.success && data.audio_base64) {
+          audioBase64 = data.audio_base64;
+          // Store in cache
+          setCachedAudio(text, audioBase64);
+          setIsLoading(false);
+        } else {
+          console.error('Failed to retrieve audio.');
+          setIsLoading(false);
+          return;
+        }
+      }
+
+      const audio = new Audio(`data:audio/mp3;base64,${audioBase64}`);
+      audioRef.current = audio;
+
+      currentlyPlayingAudio = {
+        audio,
+        stopCallback: () => {
+          setIsSpeaking(false);
+          audioRef.current = null;
+        },
+      };
+
+      audio.play().then(() => {
+        setIsSpeaking(true);
+        setIsLoading(false);
+
+        audio.onended = () => {
+          setIsSpeaking(false);
+          audioRef.current = null;
+          if (currentlyPlayingAudio?.audio === audio) {
+            currentlyPlayingAudio = null;
+          }
+        };
+      });
+    } catch (error: any) {
+      abortControllerRef.current = null;
+      currentLoadingRequest = null;
+
+      if (error.name === 'AbortError') {
+        return;
       }
-    } catch (error) {
       console.error('Error fetching audio from TTS endpoint', error);
       setIsLoading(false);
     }
   };
 
   return (
-    <div
-      className={`flex items-center justify-center rounded-full p-2 ${
-        isSpeakHovered
-          ? `dark:bg-purple-taupe bg-[#EEEEEE]`
-          : `bg-[${colorLight ? colorLight : '#FFFFFF'}] dark:bg-[${colorDark ? colorDark : 'transparent'}]`
+    <button
+      type="button"
+      className={`flex cursor-pointer items-center justify-center rounded-full p-2 ${
+        isSpeaking || isLoading
+          ? 'dark:bg-purple-taupe bg-[#EEEEEE]'
+          : 'bg-white-3000 dark:hover:bg-purple-taupe hover:bg-[#EEEEEE] dark:bg-transparent'
       }`}
+      onClick={handleSpeakClick}
+      aria-label={
+        isLoading
+          ? 'Loading audio'
+          : isSpeaking
+            ? 'Stop speaking'
+            : 'Speak text'
+      }
+      disabled={isLoading}
     >
       {isLoading ? (
         <LoadingIcon className="animate-spin" />
       ) : isSpeaking ? (
-        <Stopspeech
-          className="cursor-pointer fill-none"
-          onClick={handleSpeakClick}
-          onMouseEnter={() => setIsSpeakHovered(true)}
-          onMouseLeave={() => setIsSpeakHovered(false)}
-        />
+        <Stopspeech className="fill-none" />
       ) : (
-        <Speaker
-          className="cursor-pointer fill-none"
-          onClick={handleSpeakClick}
-          onMouseEnter={() => setIsSpeakHovered(true)}
-          onMouseLeave={() => setIsSpeakHovered(false)}
-        />
+        <Speaker className="fill-none" />
       )}
-    </div>
+    </button>
   );
 }

frontend/src/components/types/index.ts

@@ -23,6 +23,7 @@ export type InputProps = {
     e: React.KeyboardEvent<HTMLTextAreaElement | HTMLInputElement>,
   ) => void;
   leftIcon?: React.ReactNode;
+  edgeRoundness?: string;
 };
 
 export type MermaidRendererProps = {

frontend/src/conversation/Conversation.tsx

@@ -130,7 +130,7 @@ export default function Conversation() {
           }),
         );
         handleQuestion({
-          question: queries[queries.length - 1].prompt,
+          question: question,
           isRetry: true,
         });
       } else {

frontend/src/conversation/ConversationBubble.tsx

@@ -560,37 +560,47 @@ const ConversationBubble = forwardRef<
                     {handleFeedback && (
                       <>
                         <div className="relative mr-2 flex items-center justify-center">
-                          <div>
-                            <div className="bg-white-3000 dark:hover:bg-purple-taupe flex items-center justify-center rounded-full p-2 hover:bg-[#EEEEEE] dark:bg-transparent">
-                              <Like
-                                className={`${feedback === 'LIKE' ? 'fill-white-3000 stroke-purple-30 dark:fill-transparent' : 'stroke-gray-4000 fill-none'} cursor-pointer`}
-                                onClick={() => {
-                                  if (feedback === 'LIKE') {
-                                    handleFeedback?.(null);
-                                  } else {
-                                    handleFeedback?.('LIKE');
-                                  }
-                                }}
-                              ></Like>
-                            </div>
-                          </div>
+                          <button
+                            type="button"
+                            className="bg-white-3000 dark:hover:bg-purple-taupe flex cursor-pointer items-center justify-center rounded-full p-2 hover:bg-[#EEEEEE] dark:bg-transparent"
+                            onClick={() => {
+                              if (feedback === 'LIKE') {
+                                handleFeedback?.(null);
+                              } else {
+                                handleFeedback?.('LIKE');
+                              }
+                            }}
+                            aria-label={
+                              feedback === 'LIKE' ? 'Remove like' : 'Like'
+                            }
+                          >
+                            <Like
+                              className={`${feedback === 'LIKE' ? 'fill-white-3000 stroke-purple-30 dark:fill-transparent' : 'stroke-gray-4000 fill-none'}`}
+                            ></Like>
+                          </button>
                         </div>
 
                         <div className="relative mr-2 flex items-center justify-center">
-                          <div>
-                            <div className="bg-white-3000 dark:hover:bg-purple-taupe flex items-center justify-center rounded-full p-2 hover:bg-[#EEEEEE] dark:bg-transparent">
-                              <Dislike
-                                className={`${feedback === 'DISLIKE' ? 'fill-white-3000 stroke-red-2000 dark:fill-transparent' : 'stroke-gray-4000 fill-none'} cursor-pointer`}
-                                onClick={() => {
-                                  if (feedback === 'DISLIKE') {
-                                    handleFeedback?.(null);
-                                  } else {
-                                    handleFeedback?.('DISLIKE');
-                                  }
-                                }}
-                              ></Dislike>
-                            </div>
-                          </div>
+                          <button
+                            type="button"
+                            className="bg-white-3000 dark:hover:bg-purple-taupe flex cursor-pointer items-center justify-center rounded-full p-2 hover:bg-[#EEEEEE] dark:bg-transparent"
+                            onClick={() => {
+                              if (feedback === 'DISLIKE') {
+                                handleFeedback?.(null);
+                              } else {
+                                handleFeedback?.('DISLIKE');
+                              }
+                            }}
+                            aria-label={
+                              feedback === 'DISLIKE'
+                                ? 'Remove dislike'
+                                : 'Dislike'
+                            }
+                          >
+                            <Dislike
+                              className={`${feedback === 'DISLIKE' ? 'fill-white-3000 stroke-red-2000 dark:fill-transparent' : 'stroke-gray-4000 fill-none'}`}
+                            ></Dislike>
+                          </button>
                         </div>
                       </>
                     )}
@@ -793,6 +803,7 @@ function Thought({
   thought: string;
   preprocessLaTeX: (content: string) => string;
 }) {
+  const { t } = useTranslation();
   const [isDarkTheme] = useDarkTheme();
   const [isThoughtOpen, setIsThoughtOpen] = useState(true);
 
@@ -813,7 +824,9 @@ function Thought({
           className="flex flex-row items-center gap-2"
           onClick={() => setIsThoughtOpen(!isThoughtOpen)}
         >
-          <p className="text-base font-semibold">Reasoning</p>
+          <p className="text-base font-semibold">
+            {t('conversation.reasoning')}
+          </p>
           <img
             src={ChevronDown}
             alt="ChevronDown"

frontend/src/conversation/ConversationMessages.tsx

@@ -7,6 +7,7 @@ import {
   useState,
 } from 'react';
 import { useTranslation } from 'react-i18next';
+import { useSelector } from 'react-redux';
 
 import ArrowDown from '../assets/arrow-down.svg';
 import RetryIcon from '../components/RetryIcon';
@@ -14,6 +15,7 @@ import Hero from '../Hero';
 import { useDarkTheme } from '../hooks';
 import ConversationBubble from './ConversationBubble';
 import { FEEDBACK, Query, Status } from './conversationModels';
+import { selectConversationId } from '../preferences/preferenceSlice';
 
 const SCROLL_THRESHOLD = 10;
 const LAST_BUBBLE_MARGIN = 'mb-32';
@@ -50,6 +52,7 @@ export default function ConversationMessages({
 }: ConversationMessagesProps) {
   const [isDarkTheme] = useDarkTheme();
   const { t } = useTranslation();
+  const conversationId = useSelector(selectConversationId);
 
   const conversationRef = useRef<HTMLDivElement>(null);
   const [hasScrolledToLast, setHasScrolledToLast] = useState(true);
@@ -87,15 +90,20 @@ export default function ConversationMessages({
     setHasScrolledToLast(isAtBottom);
   }, [setHasScrolledToLast]);
 
+  const lastQuery = queries[queries.length - 1];
+  const lastQueryResponse = lastQuery?.response;
+  const lastQueryError = lastQuery?.error;
+  const lastQueryThought = lastQuery?.thought;
+
   useEffect(() => {
     if (!userInterruptedScroll) {
       scrollConversationToBottom();
     }
   }, [
     queries.length,
-    queries[queries.length - 1]?.response,
-    queries[queries.length - 1]?.error,
-    queries[queries.length - 1]?.thought,
+    lastQueryResponse,
+    lastQueryError,
+    lastQueryThought,
     userInterruptedScroll,
     scrollConversationToBottom,
   ]);
@@ -137,7 +145,7 @@ export default function ConversationMessages({
       return (
         <ConversationBubble
           className={bubbleMargin}
-          key={`${index}-ANSWER`}
+          key={`${conversationId}-${index}-ANSWER`}
           message={query.response}
           type={'ANSWER'}
           thought={query.thought}
@@ -175,7 +183,7 @@ export default function ConversationMessages({
       return (
         <ConversationBubble
           className={bubbleMargin}
-          key={`${index}-ERROR`}
+          key={`${conversationId}-${index}-ERROR`}
           message={query.error}
           type="ERROR"
           retryBtn={retryButton}
@@ -214,10 +222,10 @@ export default function ConversationMessages({
 
         {queries.length > 0 ? (
           queries.map((query, index) => (
-            <Fragment key={`${index}-query-fragment`}>
+            <Fragment key={`${conversationId}-${index}-query-fragment`}>
               <ConversationBubble
                 className={index === 0 ? FIRST_QUESTION_BUBBLE_MARGIN_TOP : ''}
-                key={`${index}-QUESTION`}
+                key={`${conversationId}-${index}-QUESTION`}
                 message={query.prompt}
                 type="QUESTION"
                 handleUpdatedQuestionSubmission={handleQuestionSubmission}

frontend/src/conversation/conversationHandlers.ts

@@ -15,6 +15,7 @@ export function handleFetchAnswer(
   agentId?: string,
   attachments?: string[],
   save_conversation = true,
+  modelId?: string,
 ): Promise<
   | {
       result: any;
@@ -47,6 +48,10 @@ export function handleFetchAnswer(
     save_conversation: save_conversation,
   };
 
+  if (modelId) {
+    payload.model_id = modelId;
+  }
+
   // Add attachments to payload if they exist
   if (attachments && attachments.length > 0) {
     payload.attachments = attachments;
@@ -101,6 +106,7 @@ export function handleFetchAnswerSteaming(
   agentId?: string,
   attachments?: string[],
   save_conversation = true,
+  modelId?: string,
 ): Promise<Answer> {
   const payload: RetrievalPayload = {
     question: question,
@@ -114,6 +120,10 @@ export function handleFetchAnswerSteaming(
     save_conversation: save_conversation,
   };
 
+  if (modelId) {
+    payload.model_id = modelId;
+  }
+
   // Add attachments to payload if they exist
   if (attachments && attachments.length > 0) {
     payload.attachments = attachments;

frontend/src/conversation/conversationModels.ts

@@ -65,4 +65,5 @@ export interface RetrievalPayload {
   agent_id?: string;
   attachments?: string[];
   save_conversation?: boolean;
+  model_id?: string;
 }

frontend/src/conversation/conversationSlice.ts

@@ -49,6 +49,9 @@ export const fetchAnswer = createAsyncThunk<
   }
 
   const currentConversationId = state.conversation.conversationId;
+  const modelId =
+    state.preference.selectedAgent?.default_model_id ||
+    state.preference.selectedModel?.id;
 
   if (state.preference) {
     if (API_STREAMING) {
@@ -156,7 +159,8 @@ export const fetchAnswer = createAsyncThunk<
         indx,
         state.preference.selectedAgent?.id,
         attachmentIds,
-        true, // Always save conversation
+        true,
+        modelId,
       );
     } else {
       const answer = await handleFetchAnswer(
@@ -170,7 +174,8 @@ export const fetchAnswer = createAsyncThunk<
         state.preference.token_limit,
         state.preference.selectedAgent?.id,
         attachmentIds,
-        true, // Always save conversation
+        true,
+        modelId,
       );
       if (answer) {
         let sourcesPrepped = [];
@@ -370,7 +375,10 @@ export const conversationSlice = createSlice({
           return state;
         }
         state.status = 'failed';
-        state.queries[state.queries.length - 1].error = 'Something went wrong';
+        if (state.queries.length > 0) {
+          state.queries[state.queries.length - 1].error =
+            'Something went wrong';
+        }
       });
   },
 });