Merge branch 'main' of https://github.com/arc53/DocsGPT

application/agents/base.py

@@ -1,9 +1,12 @@
+import logging
 import uuid
 from abc import ABC, abstractmethod
 from typing import Dict, Generator, List, Optional
 
 from bson.objectid import ObjectId
 
+logger = logging.getLogger(__name__)
+
 from application.agents.tools.tool_action_parser import ToolActionParser
 from application.agents.tools.tool_manager import ToolManager
 
@@ -139,6 +142,40 @@ class BaseAgent(ABC):
         tool_id, action_name, call_args = parser.parse_args(call)
 
         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)
+            
+            tool_call_data = {
+                "tool_name": "unknown",
+                "call_id": call_id,
+                "action_name": getattr(call, 'name', 'unknown'),
+                "arguments": call_args or {},
+                "result": f"Failed to parse tool call. Invalid tool name format: {getattr(call, 'name', 'unknown')}",
+            }
+            yield {"type": "tool_call", "data": {**tool_call_data, "status": "error"}}
+            self.tool_calls.append(tool_call_data)
+            return f"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,
+                "action_name": f"{action_name}_{tool_id}",
+                "arguments": call_args,
+                "result": f"Tool with ID {tool_id} not found. Available tools: {list(tools_dict.keys())}",
+            }
+            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,

application/agents/tools/tool_action_parser.py

@@ -19,8 +19,20 @@ class ToolActionParser:
     def _parse_openai_llm(self, call):
         try:
             call_args = json.loads(call.arguments)
-            tool_id = call.name.split("_")[-1]
-            action_name = call.name.rsplit("_", 1)[0]
+            tool_parts = call.name.split("_")
+            
+            # If the tool name doesn't contain an underscore, it's likely a hallucinated tool
+            if len(tool_parts) < 2:
+                logger.warning(f"Invalid tool name format: {call.name}. Expected format: action_name_tool_id")
+                return None, None, None
+                
+            tool_id = tool_parts[-1]
+            action_name = "_".join(tool_parts[:-1])
+            
+            # Validate that tool_id looks like a numerical ID
+            if not tool_id.isdigit():
+                logger.warning(f"Tool ID '{tool_id}' is not numerical. This might be a hallucinated tool call.")
+                
         except (AttributeError, TypeError) as e:
             logger.error(f"Error parsing OpenAI LLM call: {e}")
             return None, None, None
@@ -29,8 +41,20 @@ class ToolActionParser:
     def _parse_google_llm(self, call):
         try:
             call_args = call.arguments
-            tool_id = call.name.split("_")[-1]
-            action_name = call.name.rsplit("_", 1)[0]
+            tool_parts = call.name.split("_")
+            
+            # If the tool name doesn't contain an underscore, it's likely a hallucinated tool
+            if len(tool_parts) < 2:
+                logger.warning(f"Invalid tool name format: {call.name}. Expected format: action_name_tool_id")
+                return None, None, None
+                
+            tool_id = tool_parts[-1]
+            action_name = "_".join(tool_parts[:-1])
+            
+            # Validate that tool_id looks like a numerical ID
+            if not tool_id.isdigit():
+                logger.warning(f"Tool ID '{tool_id}' is not numerical. This might be a hallucinated tool call.")
+                
         except (AttributeError, TypeError) as e:
             logger.error(f"Error parsing Google LLM call: {e}")
             return None, None, None

application/parser/embedding_pipeline.py

@@ -6,6 +6,21 @@ from application.core.settings import settings
 from application.vectorstore.vector_creator import VectorCreator
 
 
+def sanitize_content(content: str) -> str:
+    """
+    Remove NUL characters that can cause vector store ingestion to fail.
+    
+    Args:
+        content (str): Raw content that may contain NUL characters
+        
+    Returns:
+        str: Sanitized content with NUL characters removed
+    """
+    if not content:
+        return content
+    return content.replace('\x00', '')
+
+
 @retry(tries=10, delay=60)
 def add_text_to_store_with_retry(store, doc, source_id):
     """
@@ -16,6 +31,9 @@ def add_text_to_store_with_retry(store, doc, source_id):
         source_id: Unique identifier for the source.
     """
     try:
+        # Sanitize content to remove NUL characters that cause ingestion failures
+        doc.page_content = sanitize_content(doc.page_content)
+        
         doc.metadata["source_id"] = str(source_id)
         store.add_texts([doc.page_content], metadatas=[doc.metadata])
     except Exception as e: