Cleanup. Text formatting. Fallback picture annotation.

Revamp UI to SSR.
Signed-off-by: DKL <dkl@zurich.ibm.com>
2025-11-29 08:33:50 +00:00 · 2025-11-24 15:17:39 +01:00 · 2025-11-21 16:15:36 +01:00 · 2025-11-21 10:31:56 +01:00 · 2025-11-20 17:57:10 +01:00 · 2025-11-17 08:31:44 +01:00
35 changed files with 9468 additions and 4054 deletions
--- a/.github/styles/config/vocabularies/Docling/accept.txt
+++ b/.github/styles/config/vocabularies/Docling/accept.txt
@@ -4,6 +4,7 @@ asgi
 async
 (?i)urls
 uvicorn
+Config
 [Ww]ebserver
 RQ
 (?i)url
--- a/.github/workflows/discord-release.yml
+++ b/.github/workflows/discord-release.yml
@@ -0,0 +1,42 @@
+# .github/workflows/discord-release.yml
+name: Notify Discord on Release
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  discord:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Send release info to Discord
+        env:
+          DISCORD_WEBHOOK: ${{ secrets.RELEASES_DISCORD_WEBHOOK }}
+        run: |
+          REPO_NAME=${{ github.repository }}
+          RELEASE_TAG=${{ github.event.release.tag_name }}
+          RELEASE_NAME="${{ github.event.release.name }}"
+          RELEASE_URL=${{ github.event.release.html_url }}
+
+          # Capture the body safely (handles backticks, $, ", etc.)
+          RELEASE_BODY=$(cat <<'EOF'
+            ${{ github.event.release.body }}
+          EOF
+          )
+
+          # Fallback if release name is empty
+          if [ -z "$RELEASE_NAME" ]; then
+            RELEASE_NAME=$RELEASE_TAG
+          fi
+
+          PAYLOAD=$(jq -n \
+          --arg title "🚀 New Release: $RELEASE_NAME" \
+          --arg url "$RELEASE_URL" \
+          --arg desc "$RELEASE_BODY" \
+          --arg author_name "$REPO_NAME" \
+          --arg author_icon "https://github.com/docling-project.png" \
+          '{embeds: [{title: $title, url: $url, description: $desc, color: 5814783, author: {name: $author_name, icon_url: $author_icon}}]}')
+
+          curl -H "Content-Type: application/json" \
+               -d "$PAYLOAD" \
+               "$DISCORD_WEBHOOK"
--- a/.github/workflows/job-image.yml
+++ b/.github/workflows/job-image.yml
@@ -108,6 +108,7 @@ jobs:
          cache-to: type=gha,mode=max
          file: Containerfile
          build-args: ${{ inputs.build_args }}
+          pull: true
      ##
      ## This stage runs after the build, so it leverages all build cache
      ## 
@@ -226,7 +227,8 @@ jobs:
          cache-to: type=gha,mode=max
          file: Containerfile
          build-args: ${{ inputs.build_args }}
+          pull: true

-      - name: Remove Local Docker Images
+      - name: Remove local Docker images
        run: |
          docker image prune -af
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -7,12 +7,12 @@ repos:
      - id: ruff-format
        name: "Ruff formatter"
        args: [--config=pyproject.toml]
-        files: '^(docling_serve|tests|examples).*\.(py|ipynb)$'
+        files: '^(docling_serve|tests|examples|scripts).*\.(py|ipynb)$'
      # Run the Ruff linter.
      - id: ruff
        name: "Ruff linter"
        args: [--exit-non-zero-on-fix, --fix, --config=pyproject.toml]
-        files: '^(docling_serve|tests|examples).*\.(py|ipynb)$'
+        files: '^(docling_serve|tests|examples|scripts).*\.(py|ipynb)$'
  - repo: local
    hooks:
      - id: system
@@ -21,6 +21,15 @@ repos:
        pass_filenames: false
        language: system
        files: '\.py$'
+  - repo: local
+    hooks:
+      - id: update-docs-common-parameters
+        name: Update Documentation File
+        entry: uv run scripts/update_doc_usage.py
+        language: python
+        pass_filenames: false
+        # Fail the commit if documentation generation fails
+        require_serial: true
  - repo: https://github.com/errata-ai/vale
    rev: v3.12.0  # Use latest stable version
    hooks:
@@ -34,6 +43,6 @@ repos:
        files: \.md$
  - repo: https://github.com/astral-sh/uv-pre-commit
    # uv version, https://github.com/astral-sh/uv-pre-commit/releases
-    rev: 0.8.3
+    rev: 0.8.19
    hooks:
      - id: uv-lock
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,106 @@
+## [v1.8.0](https://github.com/docling-project/docling-serve/releases/tag/v1.8.0) - 2025-10-31
+
+### Feature
+
+* Docling with new standard pipeline with threading ([#428](https://github.com/docling-project/docling-serve/issues/428)) ([`bf132a3`](https://github.com/docling-project/docling-serve/commit/bf132a3c3e615ddbe624841ea5b3a98593c00654))
+
+### Documentation
+
+* Expand automatic docs to nested objects. More complete usage docs. ([#426](https://github.com/docling-project/docling-serve/issues/426)) ([`35319b0`](https://github.com/docling-project/docling-serve/commit/35319b0da793a2a1a434fd2b60b7632e10ecced3))
+* Add docs for docling parameters like performance and debug ([#424](https://github.com/docling-project/docling-serve/issues/424)) ([`f3957ae`](https://github.com/docling-project/docling-serve/commit/f3957aeb577097121fe9d0d21f75a50643f03369))
+
+### Docling libraries included in this release:
+- docling 2.60.0
+- docling-core 2.50.0
+- docling-ibm-models 3.10.2
+- docling-jobkit 1.8.0
+- docling-mcp 1.3.2
+- docling-parse 4.7.0
+- docling-serve 1.8.0
+
+## [v1.7.2](https://github.com/docling-project/docling-serve/releases/tag/v1.7.2) - 2025-10-30
+
+### Fix
+
+* Update locked dependencies. Docling fixes, Expose temperature parameter for vlm models ([#423](https://github.com/docling-project/docling-serve/issues/423)) ([`e9b4140`](https://github.com/docling-project/docling-serve/commit/e9b41406c4116ff79a212877ff6484a1151e144d))
+* Temporary constrain fastapi version ([#418](https://github.com/docling-project/docling-serve/issues/418)) ([`7bf2e7b`](https://github.com/docling-project/docling-serve/commit/7bf2e7b366470e0cf1c4900df7c84becd6a96991))
+
+### Docling libraries included in this release:
+- docling 2.59.0
+- docling-core 2.50.0
+- docling-ibm-models 3.10.2
+- docling-jobkit 1.7.1
+- docling-mcp 1.3.2
+- docling-parse 4.7.0
+- docling-serve 1.7.2
+
+## [v1.7.1](https://github.com/docling-project/docling-serve/releases/tag/v1.7.1) - 2025-10-22
+
+### Fix
+
+* Upgrade dependencies ([#417](https://github.com/docling-project/docling-serve/issues/417)) ([`97613a1`](https://github.com/docling-project/docling-serve/commit/97613a19748e8c152db4a0f62b5a57fca807a33a))
+* Makes task status shared across multiple instances in RQ mode, resolves #378 ([#415](https://github.com/docling-project/docling-serve/issues/415)) ([`0961f2c`](https://github.com/docling-project/docling-serve/commit/0961f2c57425859c76130da3ea8a871d65df4b26))
+* `DOCLING_SERVE_SYNC_POLL_INTERVAL` controls the synchronous polling time ([#413](https://github.com/docling-project/docling-serve/issues/413)) ([`0f274ab`](https://github.com/docling-project/docling-serve/commit/0f274ab135a9bb41accd05db3c12a9dcce220ad9))
+
+### Documentation
+
+* Generate usage.md automatically ([#340](https://github.com/docling-project/docling-serve/issues/340)) ([`9672f31`](https://github.com/docling-project/docling-serve/commit/9672f310b1bb7030af8a276f14691e46f7da0e9e))
+
+### Docling libraries included in this release:
+- docling 2.58.0
+- docling-core 2.49.0
+- docling-ibm-models 3.10.1
+- docling-jobkit 1.7.0
+- docling-mcp 1.3.2
+- docling-parse 4.7.0
+- docling-serve 1.7.1
+
+## [v1.7.0](https://github.com/docling-project/docling-serve/releases/tag/v1.7.0) - 2025-10-17
+
+### Feature
+
+* **UI:** Add auto and orcmac options in demo UI ([#408](https://github.com/docling-project/docling-serve/issues/408)) ([`f5af71e`](https://github.com/docling-project/docling-serve/commit/f5af71e8f6de00d7dd702471a3eea2e94d882410))
+* Docling with auto-ocr ([#403](https://github.com/docling-project/docling-serve/issues/403)) ([`d95ea94`](https://github.com/docling-project/docling-serve/commit/d95ea940870af0d8df689061baa50f6026efce28))
+
+### Fix
+
+* Run docling ui behind a reverse proxy using a context path ([#396](https://github.com/docling-project/docling-serve/issues/396)) ([`5344505`](https://github.com/docling-project/docling-serve/commit/53445057184aa731ee7456b33b70bc0ecf82f2a6))
+
+### Docling libraries included in this release:
+- docling 2.57.0
+- docling-core 2.48.4
+- docling-ibm-models 3.9.1
+- docling-jobkit 1.6.0
+- docling-mcp 1.3.2
+- docling-parse 4.5.0
+- docling-serve 1.7.0
+
+## [v1.6.0](https://github.com/docling-project/docling-serve/releases/tag/v1.6.0) - 2025-10-03
+
+### Feature
+
+* Pin new version of jobkit with granite-docling and connectors ([#391](https://github.com/docling-project/docling-serve/issues/391)) ([`0595d31`](https://github.com/docling-project/docling-serve/commit/0595d31d5b357553426215ca6771796a47e41324))
+
+### Fix
+
+* Update locked dependencies ([#392](https://github.com/docling-project/docling-serve/issues/392)) ([`45f0f3c`](https://github.com/docling-project/docling-serve/commit/45f0f3c8f95d418ac30e3744d27d02a63f9e4490))
+* **UI:** Allow both lowercase and uppercase extensions ([#386](https://github.com/docling-project/docling-serve/issues/386)) ([`8b22a39`](https://github.com/docling-project/docling-serve/commit/8b22a391418d22c1a4d706f880341f28702057b5))
+* Correctly raise HTTPException for Gateway Timeout ([#382](https://github.com/docling-project/docling-serve/issues/382)) ([`d4eac05`](https://github.com/docling-project/docling-serve/commit/d4eac053f9ce0a60f9070127335bdd56e193d7fa))
+* Pinning of higher version of dependencies to fix potential security issues ([#363](https://github.com/docling-project/docling-serve/issues/363)) ([`ba61af2`](https://github.com/docling-project/docling-serve/commit/ba61af23591eff200481aa2e532cf7d0701f0ea4))
+
+### Documentation
+
+* Fix docs for websocket breaking condition ([#390](https://github.com/docling-project/docling-serve/issues/390)) ([`f6b5f0e`](https://github.com/docling-project/docling-serve/commit/f6b5f0e06354d2db7d03d274b114499e3407dccf))
+
+### Docling libraries included in this release:
+- docling 2.55.1
+- docling-core 2.48.4
+- docling-ibm-models 3.9.1
+- docling-jobkit 1.6.0
+- docling-mcp 1.3.2
+- docling-parse 4.5.0
+- docling-serve 1.6.0
+
 ## [v1.5.1](https://github.com/docling-project/docling-serve/releases/tag/v1.5.1) - 2025-09-17

 ### Fix
--- a/6
+++ b/6
@@ -1,6 +1,6 @@
 ARG BASE_IMAGE=quay.io/sclorg/python-312-c9s:c9s

-ARG UV_VERSION=0.8.3
+ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.8.19

 ARG UV_SYNC_EXTRA_ARGS=""

@@ -25,7 +25,7 @@ RUN /usr/bin/fix-permissions /opt/app-root/src/.cache

 ENV TESSDATA_PREFIX=/usr/share/tesseract/tessdata/

-FROM ghcr.io/astral-sh/uv:${UV_VERSION} AS uv_stage
+FROM ${UV_IMAGE} AS uv_stage

 ###################################################################################################
 # Docling layer                                                                                   #
@@ -58,7 +58,7 @@ RUN --mount=from=uv_stage,source=/uv,target=/bin/uv \
    uv sync ${UV_SYNC_ARGS} ${UV_SYNC_EXTRA_ARGS} --no-extra flash-attn && \
    FLASH_ATTENTION_SKIP_CUDA_BUILD=TRUE uv sync ${UV_SYNC_ARGS} ${UV_SYNC_EXTRA_ARGS} --no-build-isolation-package=flash-attn

-ARG MODELS_LIST="layout tableformer picture_classifier easyocr"
+ARG MODELS_LIST="layout tableformer picture_classifier rapidocr easyocr"

 RUN echo "Downloading models..." && \
    HF_HUB_DOWNLOAD_TIMEOUT="90" \
--- a/61
+++ b/61
@@ -16,6 +16,9 @@ else
    PIPE_DEV_NULL=
 endif

+# Container runtime - can be overridden: make CONTAINER_RUNTIME=podman cmd
+CONTAINER_RUNTIME ?= docker
+
 TAG=$(shell git rev-parse HEAD)
 BRANCH_TAG=$(shell git rev-parse --abbrev-ref HEAD)

@@ -28,44 +31,44 @@ md-lint-file:
 .PHONY: docling-serve-image
 docling-serve-image: Containerfile ## Build docling-serve container image
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve]"
-	$(CMD_PREFIX) docker build --load -f Containerfile -t ghcr.io/docling-project/docling-serve:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve:$(TAG) ghcr.io/docling-project/docling-serve:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve:$(TAG) quay.io/docling-project/docling-serve:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load -f Containerfile -t ghcr.io/docling-project/docling-serve:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve:$(TAG) ghcr.io/docling-project/docling-serve:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve:$(TAG) quay.io/docling-project/docling-serve:$(BRANCH_TAG)

 .PHONY: docling-serve-cpu-image
 docling-serve-cpu-image: Containerfile ## Build docling-serve "cpu only" container image
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve CPU]"
-	$(CMD_PREFIX) docker build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cpu --no-extra flash-attn" -f Containerfile -t ghcr.io/docling-project/docling-serve-cpu:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cpu:$(TAG) ghcr.io/docling-project/docling-serve-cpu:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cpu:$(TAG) quay.io/docling-project/docling-serve-cpu:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cpu --no-extra flash-attn" -f Containerfile -t ghcr.io/docling-project/docling-serve-cpu:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cpu:$(TAG) ghcr.io/docling-project/docling-serve-cpu:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cpu:$(TAG) quay.io/docling-project/docling-serve-cpu:$(BRANCH_TAG)

 .PHONY: docling-serve-cu124-image
 docling-serve-cu124-image: Containerfile ## Build docling-serve container image with CUDA 12.4 support
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve with Cuda 12.4]"
-	$(CMD_PREFIX) docker build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu124" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu124:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu124:$(TAG) ghcr.io/docling-project/docling-serve-cu124:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu124:$(TAG) quay.io/docling-project/docling-serve-cu124:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu124" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu124:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu124:$(TAG) ghcr.io/docling-project/docling-serve-cu124:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu124:$(TAG) quay.io/docling-project/docling-serve-cu124:$(BRANCH_TAG)

 .PHONY: docling-serve-cu126-image
 docling-serve-cu126-image: Containerfile ## Build docling-serve container image with CUDA 12.6 support
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve with Cuda 12.6]"
-	$(CMD_PREFIX) docker build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu126" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu126:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu126:$(TAG) ghcr.io/docling-project/docling-serve-cu126:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu126:$(TAG) quay.io/docling-project/docling-serve-cu126:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu126" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu126:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu126:$(TAG) ghcr.io/docling-project/docling-serve-cu126:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu126:$(TAG) quay.io/docling-project/docling-serve-cu126:$(BRANCH_TAG)

 .PHONY: docling-serve-cu128-image
 docling-serve-cu128-image: Containerfile ## Build docling-serve container image with CUDA 12.8 support
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve with Cuda 12.8]"
-	$(CMD_PREFIX) docker build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu128" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu128:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu128:$(TAG) ghcr.io/docling-project/docling-serve-cu128:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-cu128:$(TAG) quay.io/docling-project/docling-serve-cu128:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group cu128" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-cu128:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu128:$(TAG) ghcr.io/docling-project/docling-serve-cu128:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-cu128:$(TAG) quay.io/docling-project/docling-serve-cu128:$(BRANCH_TAG)

 .PHONY: docling-serve-rocm-image
 docling-serve-rocm-image: Containerfile ## Build docling-serve container image with ROCm support
 	$(ECHO_PREFIX) printf "  %-12s Containerfile\n" "[docling-serve with ROCm 6.3]"
-	$(CMD_PREFIX) docker build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group rocm --no-extra flash-attn" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-rocm:$(TAG) .
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-rocm:$(TAG) ghcr.io/docling-project/docling-serve-rocm:$(BRANCH_TAG)
-	$(CMD_PREFIX) docker tag ghcr.io/docling-project/docling-serve-rocm:$(TAG) quay.io/docling-project/docling-serve-rocm:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) build --load --build-arg "UV_SYNC_EXTRA_ARGS=--no-group pypi --group rocm --no-extra flash-attn" -f Containerfile --platform linux/amd64 -t ghcr.io/docling-project/docling-serve-rocm:$(TAG) .
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-rocm:$(TAG) ghcr.io/docling-project/docling-serve-rocm:$(BRANCH_TAG)
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) tag ghcr.io/docling-project/docling-serve-rocm:$(TAG) quay.io/docling-project/docling-serve-rocm:$(BRANCH_TAG)

 .PHONY: action-lint
 action-lint: .action-lint ##      Lint GitHub Action workflows
@@ -88,7 +91,7 @@ action-lint: .action-lint ##      Lint GitHub Action workflows
 md-lint: .md-lint ##      Lint markdown files
 .md-lint: $(wildcard */**/*.md) | md-lint-file
 	$(ECHO_PREFIX) printf "  %-12s ./...\n" "[MD LINT]"
-	$(CMD_PREFIX) docker run --rm -v $$(pwd):/workdir davidanson/markdownlint-cli2:v0.16.0 "**/*.md" "#.venv"
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run --rm -v $$(pwd):/workdir davidanson/markdownlint-cli2:v0.16.0 "**/*.md" "#.venv"
 	$(CMD_PREFIX) touch $@

 .PHONY: py-Lint
@@ -104,34 +107,34 @@ py-lint: ##      Lint Python files
 .PHONY: run-docling-cpu
 run-docling-cpu: ## Run the docling-serve container with CPU support and assign a container name
 	$(ECHO_PREFIX) printf "  %-12s Removing existing container if it exists...\n" "[CLEANUP]"
-	$(CMD_PREFIX) docker rm -f docling-serve-cpu 2>/dev/null || true
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) rm -f docling-serve-cpu 2>/dev/null || true
 	$(ECHO_PREFIX) printf "  %-12s Running docling-serve container with CPU support on port 5001...\n" "[RUN CPU]"
-	$(CMD_PREFIX) docker run -it --name docling-serve-cpu -p 5001:5001 ghcr.io/docling-project/docling-serve-cpu:main
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run -it --name docling-serve-cpu -p 5001:5001 ghcr.io/docling-project/docling-serve-cpu:main

 .PHONY: run-docling-cu124
 run-docling-cu124: ## Run the docling-serve container with GPU support and assign a container name
 	$(ECHO_PREFIX) printf "  %-12s Removing existing container if it exists...\n" "[CLEANUP]"
-	$(CMD_PREFIX) docker rm -f docling-serve-cu124 2>/dev/null || true
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) rm -f docling-serve-cu124 2>/dev/null || true
 	$(ECHO_PREFIX) printf "  %-12s Running docling-serve container with GPU support on port 5001...\n" "[RUN CUDA 12.4]"
-	$(CMD_PREFIX) docker run -it --name docling-serve-cu124 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu124:main
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run -it --name docling-serve-cu124 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu124:main

 .PHONY: run-docling-cu126
 run-docling-cu126: ## Run the docling-serve container with GPU support and assign a container name
 	$(ECHO_PREFIX) printf "  %-12s Removing existing container if it exists...\n" "[CLEANUP]"
-	$(CMD_PREFIX) docker rm -f docling-serve-cu126 2>/dev/null || true
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) rm -f docling-serve-cu126 2>/dev/null || true
 	$(ECHO_PREFIX) printf "  %-12s Running docling-serve container with GPU support on port 5001...\n" "[RUN CUDA 12.6]"
-	$(CMD_PREFIX) docker run -it --name docling-serve-cu126 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu126:main
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run -it --name docling-serve-cu126 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu126:main

 .PHONY: run-docling-cu128
 run-docling-cu128: ## Run the docling-serve container with GPU support and assign a container name
 	$(ECHO_PREFIX) printf "  %-12s Removing existing container if it exists...\n" "[CLEANUP]"
-	$(CMD_PREFIX) docker rm -f docling-serve-cu128 2>/dev/null || true
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) rm -f docling-serve-cu128 2>/dev/null || true
 	$(ECHO_PREFIX) printf "  %-12s Running docling-serve container with GPU support on port 5001...\n" "[RUN CUDA 12.8]"
-	$(CMD_PREFIX) docker run -it --name docling-serve-cu128 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu128:main
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run -it --name docling-serve-cu128 -p 5001:5001 ghcr.io/docling-project/docling-serve-cu128:main

 .PHONY: run-docling-rocm
 run-docling-rocm: ## Run the docling-serve container with GPU support and assign a container name
 	$(ECHO_PREFIX) printf "  %-12s Removing existing container if it exists...\n" "[CLEANUP]"
-	$(CMD_PREFIX) docker rm -f docling-serve-rocm 2>/dev/null || true
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) rm -f docling-serve-rocm 2>/dev/null || true
 	$(ECHO_PREFIX) printf "  %-12s Running docling-serve container with GPU support on port 5001...\n" "[RUN ROCm 6.3]"
-	$(CMD_PREFIX) docker run -it --name docling-serve-rocm -p 5001:5001 ghcr.io/docling-project/docling-serve-rocm:main
+	$(CMD_PREFIX) $(CONTAINER_RUNTIME) run -it --name docling-serve-rocm -p 5001:5001 ghcr.io/docling-project/docling-serve-rocm:main
--- a/docling_serve/main.py
+++ b/docling_serve/main.py
@@ -30,7 +30,7 @@ logger = logging.getLogger(__name__)

 def version_callback(value: bool) -> None:
    if value:
-        docling_serve_version = importlib.metadata.version("docling_serve")
+        docling_serve_version = importlib.metadata.version("docling-serve")
        docling_jobkit_version = importlib.metadata.version("docling-jobkit")
        docling_version = importlib.metadata.version("docling")
        docling_core_version = importlib.metadata.version("docling-core")
@@ -385,6 +385,11 @@ def rq_worker() -> Any:
        allow_external_plugins=docling_serve_settings.allow_external_plugins,
        max_num_pages=docling_serve_settings.max_num_pages,
        max_file_size=docling_serve_settings.max_file_size,
+        queue_max_size=docling_serve_settings.queue_max_size,
+        ocr_batch_size=docling_serve_settings.ocr_batch_size,
+        layout_batch_size=docling_serve_settings.layout_batch_size,
+        table_batch_size=docling_serve_settings.table_batch_size,
+        batch_polling_interval_seconds=docling_serve_settings.batch_polling_interval_seconds,
    )

    run_worker(
--- a/docling_serve/app.py
+++ b/docling_serve/app.py
@@ -46,7 +46,6 @@ from docling_jobkit.datamodel.s3_coords import S3Coordinates
 from docling_jobkit.datamodel.task import Task, TaskSource, TaskType
 from docling_jobkit.datamodel.task_targets import (
    InBodyTarget,
-    TaskTarget,
    ZipTarget,
 )
 from docling_jobkit.orchestrators.base_orchestrator import (
@@ -55,7 +54,7 @@ from docling_jobkit.orchestrators.base_orchestrator import (
    TaskNotFoundError,
 )

-from docling_serve.auth import APIKeyAuth, AuthenticationResult
+from docling_serve.auth import APIKeyHeaderAuth, AuthenticationResult
 from docling_serve.datamodel.convert import ConvertDocumentsRequestOptions
 from docling_serve.datamodel.requests import (
    ConvertDocumentsRequest,
@@ -64,6 +63,7 @@ from docling_serve.datamodel.requests import (
    HttpSourceRequest,
    S3SourceRequest,
    TargetName,
+    TargetRequest,
    make_request_model,
 )
 from docling_serve.datamodel.responses import (
@@ -76,7 +76,7 @@ from docling_serve.datamodel.responses import (
    TaskStatusResponse,
    WebsocketMessage,
 )
-from docling_serve.helper_functions import FormDepends
+from docling_serve.helper_functions import DOCLING_VERSIONS, FormDepends
 from docling_serve.orchestrator_factory import get_async_orchestrator
 from docling_serve.response_preparation import prepare_response
 from docling_serve.settings import docling_serve_settings
@@ -167,7 +167,7 @@ def create_app():  # noqa: C901
        offline_docs_assets = True
        _log.info("Found static assets.")

-    require_auth = APIKeyAuth(docling_serve_settings.api_key)
+    require_auth = APIKeyHeaderAuth(docling_serve_settings.api_key)
    app = FastAPI(
        title="Docling Serve",
        docs_url=None if offline_docs_assets else "/swagger",
@@ -188,30 +188,6 @@ def create_app():  # noqa: C901
        allow_headers=headers,
    )

-    # Mount the Gradio app
-    if docling_serve_settings.enable_ui:
-        try:
-            import gradio as gr
-
-            from docling_serve.gradio_ui import ui as gradio_ui
-
-            tmp_output_dir = get_scratch() / "gradio"
-            tmp_output_dir.mkdir(exist_ok=True, parents=True)
-            gradio_ui.gradio_output_dir = tmp_output_dir
-            app = gr.mount_gradio_app(
-                app,
-                gradio_ui,
-                path="/ui",
-                allowed_paths=["./logo.png", tmp_output_dir],
-                root_path="/ui",
-            )
-        except ImportError:
-            _log.warning(
-                "Docling Serve enable_ui is activated, but gradio is not installed. "
-                "Install it with `pip install docling-serve[ui]` "
-                "or `pip install gradio`"
-            )
-
    #############################
    # Offline assets definition #
    #############################
@@ -304,7 +280,7 @@ def create_app():  # noqa: C901
        convert_options: ConvertDocumentsRequestOptions,
        chunking_options: BaseChunkerOptions | None,
        chunking_export_options: ChunkingExportOptions | None,
-        target: TaskTarget,
+        target: TargetRequest,
    ) -> Task:
        _log.info(f"Received {len(files)} files for processing.")

@@ -332,7 +308,7 @@ def create_app():  # noqa: C901
            task = await orchestrator.task_status(task_id=task_id)
            if task.is_completed():
                return True
-            await asyncio.sleep(5)
+            await asyncio.sleep(docling_serve_settings.sync_poll_interval)
            elapsed_time = time.monotonic() - start_time
            if elapsed_time > docling_serve_settings.max_sync_wait:
                return False
@@ -428,6 +404,16 @@ def create_app():  # noqa: C901
    def api_check() -> HealthCheckResponse:
        return HealthCheckResponse()

+    # Docling versions
+    @app.get("/version", tags=["health"])
+    def version_info() -> dict:
+        if not docling_serve_settings.show_version_info:
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="Forbidden. The server is configured for not showing version details.",
+            )
+        return DOCLING_VERSIONS
+
    # Convert a document from URL(s)
    @app.post(
        "/v1/convert/source",
@@ -455,7 +441,7 @@ def create_app():  # noqa: C901

        if not completed:
            # TODO: abort task!
-            return HTTPException(
+            raise HTTPException(
                status_code=504,
                detail=f"Conversion is taking too long. The maximum wait time is configure as DOCLING_SERVE_MAX_SYNC_WAIT={docling_serve_settings.max_sync_wait}.",
            )
@@ -511,7 +497,7 @@ def create_app():  # noqa: C901

        if not completed:
            # TODO: abort task!
-            return HTTPException(
+            raise HTTPException(
                status_code=504,
                detail=f"Conversion is taking too long. The maximum wait time is configure as DOCLING_SERVE_MAX_SYNC_WAIT={docling_serve_settings.max_sync_wait}.",
            )
@@ -711,7 +697,7 @@ def create_app():  # noqa: C901

            if not completed:
                # TODO: abort task!
-                return HTTPException(
+                raise HTTPException(
                    status_code=504,
                    detail=f"Conversion is taking too long. The maximum wait time is configure as DOCLING_SERVE_MAX_SYNC_WAIT={docling_serve_settings.max_sync_wait}.",
                )
@@ -793,7 +779,7 @@ def create_app():  # noqa: C901

            if not completed:
                # TODO: abort task!
-                return HTTPException(
+                raise HTTPException(
                    status_code=504,
                    detail=f"Conversion is taking too long. The maximum wait time is configure as DOCLING_SERVE_MAX_SYNC_WAIT={docling_serve_settings.max_sync_wait}.",
                )
@@ -860,7 +846,10 @@ def create_app():  # noqa: C901
        assert isinstance(orchestrator.notifier, WebsocketNotifier)
        await websocket.accept()

-        if task_id not in orchestrator.tasks:
+        try:
+            # Get task status from Redis or RQ directly instead of checking in-memory registry
+            task = await orchestrator.task_status(task_id=task_id)
+        except TaskNotFoundError:
            await websocket.send_text(
                WebsocketMessage(
                    message=MessageKind.ERROR, error="Task not found."
@@ -869,8 +858,6 @@ def create_app():  # noqa: C901
            await websocket.close()
            return

-        task = orchestrator.tasks[task_id]
-
        # Track active WebSocket connections for this job
        orchestrator.notifier.task_subscribers[task_id].add(websocket)

@@ -1001,4 +988,20 @@ def create_app():  # noqa: C901
        await orchestrator.clear_results(older_than=older_then)
        return ClearResponse()

+    # Optional UI
+    if docling_serve_settings.enable_ui:
+        try:
+            from docling_serve.ui.app import create_ui_app
+
+            ui_app = create_ui_app(
+                process_file_async, process_url_async, task_result, task_status_poll
+            )
+            app.mount("/ui", app=ui_app, name="ui")
+        except ImportError as ex:
+            _log.error(ex)
+            _log.warning(
+                "Docling Serve enable_ui is activated, but its dependencies are not installed."
+                "Install it with `uv sync --extra ui`"
+            )
+
    return app
--- a/docling_serve/auth.py
+++ b/docling_serve/auth.py
@@ -1,7 +1,7 @@
 from typing import Any

-from fastapi import HTTPException, Request, status
-from fastapi.security import APIKeyHeader
+from fastapi import HTTPException, Request, Response, status
+from fastapi.security import APIKeyCookie, APIKeyHeader
 from pydantic import BaseModel


@@ -11,46 +11,79 @@ class AuthenticationResult(BaseModel):
    detail: Any | None = None


-class APIKeyAuth(APIKeyHeader):
-    """
-    FastAPI dependency which evaluates a status API Key.
-    """
-
+class KeyValidator:
    def __init__(
        self,
        api_key: str,
-        header_name: str = "X-Api-Key",
+        field_name: str = "X-Api-Key",
        fail_on_unauthorized: bool = True,
    ) -> None:
        self.api_key = api_key
-        self.header_name = header_name
-        super().__init__(name=self.header_name, auto_error=False)
+        self.field_name = field_name
+        self.fail_on_unauthorized = fail_on_unauthorized

-    async def _validate_api_key(self, header_api_key: str | None):
-        if header_api_key is None:
-            return AuthenticationResult(
-                valid=False, errors=[f"Missing header {self.header_name}."]
-            )
+    async def __call__(self, candidate_key: str | None):
+        if candidate_key is None:
+            return self._error(f"Missing field {self.field_name}.")

-        header_api_key = header_api_key.strip()
+        candidate_key = candidate_key.strip()

        # Otherwise check the apikey
-        if header_api_key == self.api_key or self.api_key == "":
+        if candidate_key == self.api_key or self.api_key == "":
            return AuthenticationResult(
                valid=True,
-                detail=header_api_key,
+                detail=candidate_key,  # Remove?
            )
+        else:
+            return self._error("The provided API Key is invalid.")
+
+    def _error(self, error: str):
+        if self.fail_on_unauthorized and self.api_key:
+            raise HTTPException(status.HTTP_401_UNAUTHORIZED, error)
        else:
            return AuthenticationResult(
                valid=False,
-                errors=["The provided API Key is invalid."],
+                errors=[error],
            )

+
+class APIKeyHeaderAuth(APIKeyHeader):
+    """
+    FastAPI dependency which evaluates a status API Key in a header.
+    """
+
+    def __init__(self, validator: str | KeyValidator) -> None:
+        self.validator = (
+            KeyValidator(validator) if isinstance(validator, str) else validator
+        )
+        super().__init__(name=self.validator.field_name, auto_error=False)
+
    async def __call__(self, request: Request) -> AuthenticationResult:  # type: ignore
-        header_api_key = await super().__call__(request=request)
-        result = await self._validate_api_key(header_api_key)
-        if self.api_key and not result.valid:
-            raise HTTPException(
-                status_code=status.HTTP_401_UNAUTHORIZED, detail=result.detail
-            )
-        return result
+        key = await super().__call__(request=request)
+        return await self.validator(key)
+
+
+class APIKeyCookieAuth(APIKeyCookie):
+    """
+    FastAPI dependency which evaluates a status API Key in a cookie.
+    """
+
+    def __init__(self, validator: str | KeyValidator) -> None:
+        self.validator = (
+            KeyValidator(validator) if isinstance(validator, str) else validator
+        )
+        super().__init__(name=self.validator.field_name, auto_error=False)
+
+    async def __call__(self, request: Request) -> AuthenticationResult:  # type: ignore
+        api_key = await super().__call__(request=request)
+        return await self.validator(api_key)
+
+    def _set_api_key(self, response: Response, api_key: str, expires=24 * 3600):
+        response.set_cookie(
+            key=self.validator.field_name,
+            value=api_key,
+            expires=expires,
+            secure=True,
+            httponly=True,
+            samesite="strict",
+        )
--- a/docling_serve/datamodel/requests.py
+++ b/docling_serve/datamodel/requests.py
@@ -13,8 +13,8 @@ from docling_jobkit.datamodel.http_inputs import FileSource, HttpSource
 from docling_jobkit.datamodel.s3_coords import S3Coordinates
 from docling_jobkit.datamodel.task_targets import (
    InBodyTarget,
+    PutTarget,
    S3Target,
-    TaskTarget,
    ZipTarget,
 )

@@ -47,12 +47,17 @@ SourceRequestItem = Annotated[
    FileSourceRequest | HttpSourceRequest | S3SourceRequest, Field(discriminator="kind")
 ]

+TargetRequest = Annotated[
+    InBodyTarget | ZipTarget | S3Target | PutTarget,
+    Field(discriminator="kind"),
+]
+

 ## Complete Source request
 class ConvertDocumentsRequest(BaseModel):
    options: ConvertDocumentsRequestOptions = ConvertDocumentsRequestOptions()
    sources: list[SourceRequestItem]
-    target: TaskTarget = InBodyTarget()
+    target: TargetRequest = InBodyTarget()

    @model_validator(mode="after")
    def validate_s3_source_and_target(self) -> Self:
@@ -94,7 +99,7 @@ class BaseChunkDocumentsRequest(BaseModel):
        ),
    ] = False
    target: Annotated[
-        TaskTarget, Field(description="Specification for the type of output target.")
+        TargetRequest, Field(description="Specification for the type of output target.")
    ] = InBodyTarget()


--- a/docling_serve/gradio_ui.py
+++ b/docling_serve/gradio_ui.py
@@ -1,906 +0,0 @@
-import base64
-import importlib
-import itertools
-import json
-import logging
-import ssl
-import tempfile
-import time
-from pathlib import Path
-from typing import Optional
-
-import certifi
-import gradio as gr
-import httpx
-
-from docling.datamodel.base_models import FormatToExtensions
-from docling.datamodel.pipeline_options import (
-    PdfBackend,
-    ProcessingPipeline,
-    TableFormerMode,
-    TableStructureOptions,
-)
-
-from docling_serve.helper_functions import _to_list_of_strings
-from docling_serve.settings import docling_serve_settings, uvicorn_settings
-
-logger = logging.getLogger(__name__)
-
-############################
-# Path of static artifacts #
-############################
-
-logo_path = "https://raw.githubusercontent.com/docling-project/docling/refs/heads/main/docs/assets/logo.svg"
-js_components_url = "https://unpkg.com/@docling/docling-components@0.0.7"
-if (
-    docling_serve_settings.static_path is not None
-    and docling_serve_settings.static_path.is_dir()
-):
-    logo_path = str(docling_serve_settings.static_path / "logo.svg")
-    js_components_url = "/static/docling-components.js"
-
-
-##############################
-# Head JS for web components #
-##############################
-head = f"""
-    <script src="{js_components_url}" type="module"></script>
-"""
-
-#################
-# CSS and theme #
-#################
-
-css = """
-#logo {
-    border-style: none;
-    background: none;
-    box-shadow: none;
-    min-width: 80px;
-}
-#dark_mode_column {
-    display: flex;
-    align-content: flex-end;
-}
-#title {
-    text-align: left;
-    display:block;
-    height: auto;
-    padding-top: 5px;
-    line-height: 0;
-}
-.title-text h1 > p, .title-text p {
-    margin-top: 0px !important;
-    margin-bottom: 2px !important;
-}
-#custom-container {
-    border: 0.909091px solid;
-    padding: 10px;
-    border-radius: 4px;
-}
-#custom-container h4 {
-    font-size: 14px;
-}
-#file_input_zone {
-    height: 140px;
-}
-
-docling-img {
-    gap: 1rem;
-}
-
-docling-img::part(page) {
-    box-shadow: 0 0.5rem 1rem 0 rgba(0, 0, 0, 0.2);
-}
-"""
-
-theme = gr.themes.Default(
-    text_size="md",
-    spacing_size="md",
-    font=[
-        gr.themes.GoogleFont("Red Hat Display"),
-        "ui-sans-serif",
-        "system-ui",
-        "sans-serif",
-    ],
-    font_mono=[
-        gr.themes.GoogleFont("Red Hat Mono"),
-        "ui-monospace",
-        "Consolas",
-        "monospace",
-    ],
-)
-
-#############
-# Variables #
-#############
-
-gradio_output_dir = None  # Will be set by FastAPI when mounted
-file_output_path = None  # Will be set when a new file is generated
-
-#############
-# Functions #
-#############
-
-
-def get_api_endpoint() -> str:
-    protocol = "http"
-    if uvicorn_settings.ssl_keyfile is not None:
-        protocol = "https"
-    return f"{protocol}://{docling_serve_settings.api_host}:{uvicorn_settings.port}"
-
-
-def get_ssl_context() -> ssl.SSLContext:
-    ctx = ssl.create_default_context(cafile=certifi.where())
-    kube_sa_ca_cert_path = Path(
-        "/run/secrets/kubernetes.io/serviceaccount/service-ca.crt"
-    )
-    if (
-        uvicorn_settings.ssl_keyfile is not None
-        and ".svc." in docling_serve_settings.api_host
-        and kube_sa_ca_cert_path.exists()
-    ):
-        ctx.load_verify_locations(cafile=kube_sa_ca_cert_path)
-    return ctx
-
-
-def health_check():
-    response = httpx.get(f"{get_api_endpoint()}/health")
-    if response.status_code == 200:
-        return "Healthy"
-    return "Unhealthy"
-
-
-def set_options_visibility(x):
-    return gr.Accordion("Options", open=x)
-
-
-def set_outputs_visibility_direct(x, y):
-    content = gr.Row(visible=x)
-    file = gr.Row(visible=y)
-    return content, file
-
-
-def set_task_id_visibility(x):
-    task_id_row = gr.Row(visible=x)
-    return task_id_row
-
-
-def set_outputs_visibility_process(x):
-    content = gr.Row(visible=not x)
-    file = gr.Row(visible=x)
-    return content, file
-
-
-def set_download_button_label(label_text: gr.State):
-    return gr.DownloadButton(label=str(label_text), scale=1)
-
-
-def clear_outputs():
-    task_id_rendered = ""
-    markdown_content = ""
-    json_content = ""
-    json_rendered_content = ""
-    html_content = ""
-    text_content = ""
-    doctags_content = ""
-
-    return (
-        task_id_rendered,
-        markdown_content,
-        markdown_content,
-        json_content,
-        json_rendered_content,
-        html_content,
-        html_content,
-        text_content,
-        doctags_content,
-    )
-
-
-def clear_url_input():
-    return ""
-
-
-def clear_file_input():
-    return None
-
-
-def auto_set_return_as_file(
-    url_input_value: str,
-    file_input_value: Optional[list[str]],
-    image_export_mode_value: str,
-):
-    # If more than one input source is provided, return as file
-    if (
-        (len(url_input_value.split(",")) > 1)
-        or (file_input_value and len(file_input_value) > 1)
-        or (image_export_mode_value == "referenced")
-    ):
-        return True
-    else:
-        return False
-
-
-def change_ocr_lang(ocr_engine):
-    if ocr_engine == "easyocr":
-        return "en,fr,de,es"
-    elif ocr_engine == "tesseract_cli":
-        return "eng,fra,deu,spa"
-    elif ocr_engine == "tesseract":
-        return "eng,fra,deu,spa"
-    elif ocr_engine == "rapidocr":
-        return "english,chinese"
-
-
-def wait_task_finish(auth: str, task_id: str, return_as_file: bool):
-    conversion_sucess = False
-    task_finished = False
-    task_status = ""
-
-    headers = {}
-    if docling_serve_settings.api_key:
-        headers["X-Api-Key"] = str(auth)
-
-    ssl_ctx = get_ssl_context()
-    while not task_finished:
-        try:
-            response = httpx.get(
-                f"{get_api_endpoint()}/v1/status/poll/{task_id}?wait=5",
-                headers=headers,
-                verify=ssl_ctx,
-                timeout=15,
-            )
-            task_status = response.json()["task_status"]
-            if task_status == "success":
-                conversion_sucess = True
-                task_finished = True
-
-            if task_status in ("failure", "revoked"):
-                conversion_sucess = False
-                task_finished = True
-                raise RuntimeError(f"Task failed with status {task_status!r}")
-            time.sleep(5)
-        except Exception as e:
-            logger.error(f"Error processing file(s): {e}")
-            conversion_sucess = False
-            task_finished = True
-            raise gr.Error(f"Error processing file(s): {e}", print_exception=False)
-
-    if conversion_sucess:
-        try:
-            response = httpx.get(
-                f"{get_api_endpoint()}/v1/result/{task_id}",
-                headers=headers,
-                timeout=15,
-                verify=ssl_ctx,
-            )
-            output = response_to_output(response, return_as_file)
-            return output
-        except Exception as e:
-            logger.error(f"Error getting task result: {e}")
-
-    raise gr.Error(
-        f"Error getting task result, conversion finished with status: {task_status}"
-    )
-
-
-def process_url(
-    auth,
-    input_sources,
-    to_formats,
-    image_export_mode,
-    pipeline,
-    ocr,
-    force_ocr,
-    ocr_engine,
-    ocr_lang,
-    pdf_backend,
-    table_mode,
-    abort_on_error,
-    return_as_file,
-    do_code_enrichment,
-    do_formula_enrichment,
-    do_picture_classification,
-    do_picture_description,
-):
-    target = {"kind": "zip" if return_as_file else "inbody"}
-    parameters = {
-        "sources": [
-            {"kind": "http", "url": source} for source in input_sources.split(",")
-        ],
-        "options": {
-            "to_formats": to_formats,
-            "image_export_mode": image_export_mode,
-            "pipeline": pipeline,
-            "ocr": ocr,
-            "force_ocr": force_ocr,
-            "ocr_engine": ocr_engine,
-            "ocr_lang": _to_list_of_strings(ocr_lang),
-            "pdf_backend": pdf_backend,
-            "table_mode": table_mode,
-            "abort_on_error": abort_on_error,
-            "do_code_enrichment": do_code_enrichment,
-            "do_formula_enrichment": do_formula_enrichment,
-            "do_picture_classification": do_picture_classification,
-            "do_picture_description": do_picture_description,
-        },
-        "target": target,
-    }
-    if (
-        not parameters["sources"]
-        or len(parameters["sources"]) == 0
-        or parameters["sources"][0]["url"] == ""
-    ):
-        logger.error("No input sources provided.")
-        raise gr.Error("No input sources provided.", print_exception=False)
-
-    headers = {}
-    if docling_serve_settings.api_key:
-        headers["X-Api-Key"] = str(auth)
-
-    print(f"{headers=}")
-    try:
-        ssl_ctx = get_ssl_context()
-        response = httpx.post(
-            f"{get_api_endpoint()}/v1/convert/source/async",
-            json=parameters,
-            headers=headers,
-            verify=ssl_ctx,
-            timeout=60,
-        )
-    except Exception as e:
-        logger.error(f"Error processing URL: {e}")
-        raise gr.Error(f"Error processing URL: {e}", print_exception=False)
-    if response.status_code != 200:
-        data = response.json()
-        error_message = data.get("detail", "An unknown error occurred.")
-        logger.error(f"Error processing file: {error_message}")
-        raise gr.Error(f"Error processing file: {error_message}", print_exception=False)
-
-    task_id_rendered = response.json()["task_id"]
-    return task_id_rendered
-
-
-def file_to_base64(file):
-    with open(file.name, "rb") as f:
-        encoded_string = base64.b64encode(f.read()).decode("utf-8")
-    return encoded_string
-
-
-def process_file(
-    auth,
-    files,
-    to_formats,
-    image_export_mode,
-    pipeline,
-    ocr,
-    force_ocr,
-    ocr_engine,
-    ocr_lang,
-    pdf_backend,
-    table_mode,
-    abort_on_error,
-    return_as_file,
-    do_code_enrichment,
-    do_formula_enrichment,
-    do_picture_classification,
-    do_picture_description,
-):
-    if not files or len(files) == 0:
-        logger.error("No files provided.")
-        raise gr.Error("No files provided.", print_exception=False)
-    files_data = [
-        {"kind": "file", "base64_string": file_to_base64(file), "filename": file.name}
-        for file in files
-    ]
-    target = {"kind": "zip" if return_as_file else "inbody"}
-
-    parameters = {
-        "sources": files_data,
-        "options": {
-            "to_formats": to_formats,
-            "image_export_mode": image_export_mode,
-            "pipeline": pipeline,
-            "ocr": ocr,
-            "force_ocr": force_ocr,
-            "ocr_engine": ocr_engine,
-            "ocr_lang": _to_list_of_strings(ocr_lang),
-            "pdf_backend": pdf_backend,
-            "table_mode": table_mode,
-            "abort_on_error": abort_on_error,
-            "return_as_file": return_as_file,
-            "do_code_enrichment": do_code_enrichment,
-            "do_formula_enrichment": do_formula_enrichment,
-            "do_picture_classification": do_picture_classification,
-            "do_picture_description": do_picture_description,
-        },
-        "target": target,
-    }
-
-    headers = {}
-    if docling_serve_settings.api_key:
-        headers["X-Api-Key"] = str(auth)
-
-    try:
-        ssl_ctx = get_ssl_context()
-        response = httpx.post(
-            f"{get_api_endpoint()}/v1/convert/source/async",
-            json=parameters,
-            headers=headers,
-            verify=ssl_ctx,
-            timeout=60,
-        )
-    except Exception as e:
-        logger.error(f"Error processing file(s): {e}")
-        raise gr.Error(f"Error processing file(s): {e}", print_exception=False)
-    if response.status_code != 200:
-        data = response.json()
-        error_message = data.get("detail", "An unknown error occurred.")
-        logger.error(f"Error processing file: {error_message}")
-        raise gr.Error(f"Error processing file: {error_message}", print_exception=False)
-
-    task_id_rendered = response.json()["task_id"]
-    return task_id_rendered
-
-
-def response_to_output(response, return_as_file):
-    markdown_content = ""
-    json_content = ""
-    json_rendered_content = ""
-    html_content = ""
-    text_content = ""
-    doctags_content = ""
-    download_button = gr.DownloadButton(visible=False, label="Download Output", scale=1)
-    if return_as_file:
-        filename = (
-            response.headers.get("Content-Disposition").split("filename=")[1].strip('"')
-        )
-        tmp_output_dir = Path(tempfile.mkdtemp(dir=gradio_output_dir, prefix="ui_"))
-        file_output_path = f"{tmp_output_dir}/{filename}"
-        # logger.info(f"Saving file to: {file_output_path}")
-        with open(file_output_path, "wb") as f:
-            f.write(response.content)
-        download_button = gr.DownloadButton(
-            visible=True, label=f"Download {filename}", scale=1, value=file_output_path
-        )
-    else:
-        full_content = response.json()
-        markdown_content = full_content.get("document").get("md_content")
-        json_content = json.dumps(
-            full_content.get("document").get("json_content"), indent=2
-        )
-        # Embed document JSON and trigger load at client via an image.
-        json_rendered_content = f"""
-            <docling-img id="dclimg" pagenumbers><docling-tooltip></docling-tooltip></docling-img>
-            <script id="dcljson" type="application/json" onload="document.getElementById('dclimg').src = JSON.parse(document.getElementById('dcljson').textContent);">{json_content}</script>
-            <img src onerror="document.getElementById('dclimg').src = JSON.parse(document.getElementById('dcljson').textContent);" />
-            """
-        html_content = full_content.get("document").get("html_content")
-        text_content = full_content.get("document").get("text_content")
-        doctags_content = full_content.get("document").get("doctags_content")
-    return (
-        markdown_content,
-        markdown_content,
-        json_content,
-        json_rendered_content,
-        html_content,
-        html_content,
-        text_content,
-        doctags_content,
-        download_button,
-    )
-
-
-############
-# UI Setup #
-############
-
-with gr.Blocks(
-    head=head,
-    css=css,
-    theme=theme,
-    title="Docling Serve",
-    delete_cache=(3600, 36000),  # Delete all files older than 10 hour every hour
-) as ui:
-    # Constants stored in states to be able to pass them as inputs to functions
-    processing_text = gr.State("Processing your document(s), please wait...")
-    true_bool = gr.State(True)
-    false_bool = gr.State(False)
-
-    # Banner
-    with gr.Row(elem_id="check_health"):
-        # Logo
-        with gr.Column(scale=1, min_width=90):
-            try:
-                gr.Image(
-                    logo_path,
-                    height=80,
-                    width=80,
-                    show_download_button=False,
-                    show_label=False,
-                    show_fullscreen_button=False,
-                    container=False,
-                    elem_id="logo",
-                    scale=0,
-                )
-            except Exception:
-                logger.warning("Logo not found.")
-
-        # Title
-        with gr.Column(scale=1, min_width=200):
-            gr.Markdown(
-                f"# Docling Serve \n(docling version: "
-                f"{importlib.metadata.version('docling')})",
-                elem_id="title",
-                elem_classes=["title-text"],
-            )
-        # Dark mode button
-        with gr.Column(scale=16, elem_id="dark_mode_column"):
-            dark_mode_btn = gr.Button("Dark/Light Mode", scale=0)
-            dark_mode_btn.click(
-                None,
-                None,
-                None,
-                js="""() => {
-                    if (document.querySelectorAll('.dark').length) {
-                        document.querySelectorAll('.dark').forEach(
-                        el => el.classList.remove('dark')
-                        );
-                    } else {
-                        document.querySelector('body').classList.add('dark');
-                    }
-                }""",
-                show_api=False,
-            )
-
-    # URL Processing Tab
-    with gr.Tab("Convert URL"):
-        with gr.Row():
-            with gr.Column(scale=4):
-                url_input = gr.Textbox(
-                    label="URL Input Source",
-                    placeholder="https://arxiv.org/pdf/2501.17887",
-                )
-            with gr.Column(scale=1):
-                url_process_btn = gr.Button("Process URL", scale=1)
-                url_reset_btn = gr.Button("Reset", scale=1)
-
-    # File Processing Tab
-    with gr.Tab("Convert File"):
-        with gr.Row():
-            with gr.Column(scale=4):
-                file_input = gr.File(
-                    elem_id="file_input_zone",
-                    label="Upload File",
-                    file_types=[
-                        f".{v}"
-                        for v in itertools.chain.from_iterable(
-                            FormatToExtensions.values()
-                        )
-                    ],
-                    file_count="multiple",
-                    scale=4,
-                )
-            with gr.Column(scale=1):
-                file_process_btn = gr.Button("Process File", scale=1)
-                file_reset_btn = gr.Button("Reset", scale=1)
-
-    # Auth
-    with gr.Row(visible=bool(docling_serve_settings.api_key)):
-        with gr.Column():
-            auth = gr.Textbox(
-                label="Authentication",
-                placeholder="API Key",
-                type="password",
-            )
-
-    # Options
-    with gr.Accordion("Options") as options:
-        with gr.Row():
-            with gr.Column(scale=1):
-                to_formats = gr.CheckboxGroup(
-                    [
-                        ("Docling (JSON)", "json"),
-                        ("Markdown", "md"),
-                        ("HTML", "html"),
-                        ("Plain Text", "text"),
-                        ("Doc Tags", "doctags"),
-                    ],
-                    label="To Formats",
-                    value=["json", "md"],
-                )
-            with gr.Column(scale=1):
-                image_export_mode = gr.Radio(
-                    [
-                        ("Embedded", "embedded"),
-                        ("Placeholder", "placeholder"),
-                        ("Referenced", "referenced"),
-                    ],
-                    label="Image Export Mode",
-                    value="embedded",
-                )
-
-        with gr.Row():
-            with gr.Column(scale=1, min_width=200):
-                pipeline = gr.Radio(
-                    [(v.value.capitalize(), v.value) for v in ProcessingPipeline],
-                    label="Pipeline type",
-                    value=ProcessingPipeline.STANDARD.value,
-                )
-        with gr.Row():
-            with gr.Column(scale=1, min_width=200):
-                ocr = gr.Checkbox(label="Enable OCR", value=True)
-                force_ocr = gr.Checkbox(label="Force OCR", value=False)
-            with gr.Column(scale=1):
-                ocr_engine = gr.Radio(
-                    [
-                        ("EasyOCR", "easyocr"),
-                        ("Tesseract", "tesseract"),
-                        ("RapidOCR", "rapidocr"),
-                    ],
-                    label="OCR Engine",
-                    value="easyocr",
-                )
-            with gr.Column(scale=1, min_width=200):
-                ocr_lang = gr.Textbox(
-                    label="OCR Language (beware of the format)", value="en,fr,de,es"
-                )
-            ocr_engine.change(change_ocr_lang, inputs=[ocr_engine], outputs=[ocr_lang])
-        with gr.Row():
-            with gr.Column(scale=4):
-                pdf_backend = gr.Radio(
-                    [v.value for v in PdfBackend],
-                    label="PDF Backend",
-                    value=PdfBackend.DLPARSE_V4.value,
-                )
-            with gr.Column(scale=2):
-                table_mode = gr.Radio(
-                    [(v.value.capitalize(), v.value) for v in TableFormerMode],
-                    label="Table Mode",
-                    value=TableStructureOptions().mode.value,
-                )
-            with gr.Column(scale=1):
-                abort_on_error = gr.Checkbox(label="Abort on Error", value=False)
-                return_as_file = gr.Checkbox(label="Return as File", value=False)
-        with gr.Row():
-            with gr.Column():
-                do_code_enrichment = gr.Checkbox(
-                    label="Enable code enrichment", value=False
-                )
-                do_formula_enrichment = gr.Checkbox(
-                    label="Enable formula enrichment", value=False
-                )
-            with gr.Column():
-                do_picture_classification = gr.Checkbox(
-                    label="Enable picture classification", value=False
-                )
-                do_picture_description = gr.Checkbox(
-                    label="Enable picture description", value=False
-                )
-
-    # Task id output
-    with gr.Row(visible=False) as task_id_output:
-        task_id_rendered = gr.Textbox(label="Task id", interactive=False)
-
-    # Document output
-    with gr.Row(visible=False) as content_output:
-        with gr.Tab("Docling (JSON)"):
-            output_json = gr.Code(language="json", wrap_lines=True, show_label=False)
-        with gr.Tab("Docling-Rendered"):
-            output_json_rendered = gr.HTML(label="Response")
-        with gr.Tab("Markdown"):
-            output_markdown = gr.Code(
-                language="markdown", wrap_lines=True, show_label=False
-            )
-        with gr.Tab("Markdown-Rendered"):
-            output_markdown_rendered = gr.Markdown(label="Response")
-        with gr.Tab("HTML"):
-            output_html = gr.Code(language="html", wrap_lines=True, show_label=False)
-        with gr.Tab("HTML-Rendered"):
-            output_html_rendered = gr.HTML(label="Response")
-        with gr.Tab("Text"):
-            output_text = gr.Code(wrap_lines=True, show_label=False)
-        with gr.Tab("DocTags"):
-            output_doctags = gr.Code(wrap_lines=True, show_label=False)
-
-    # File download output
-    with gr.Row(visible=False) as file_output:
-        download_file_btn = gr.DownloadButton(label="Placeholder", scale=1)
-
-    ##############
-    # UI Actions #
-    ##############
-
-    # Handle Return as File
-    url_input.change(
-        auto_set_return_as_file,
-        inputs=[url_input, file_input, image_export_mode],
-        outputs=[return_as_file],
-    )
-    file_input.change(
-        auto_set_return_as_file,
-        inputs=[url_input, file_input, image_export_mode],
-        outputs=[return_as_file],
-    )
-    image_export_mode.change(
-        auto_set_return_as_file,
-        inputs=[url_input, file_input, image_export_mode],
-        outputs=[return_as_file],
-    )
-
-    # URL processing
-    url_process_btn.click(
-        set_options_visibility, inputs=[false_bool], outputs=[options]
-    ).then(
-        set_download_button_label, inputs=[processing_text], outputs=[download_file_btn]
-    ).then(
-        clear_outputs,
-        inputs=None,
-        outputs=[
-            task_id_rendered,
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-        ],
-    ).then(
-        set_task_id_visibility,
-        inputs=[true_bool],
-        outputs=[task_id_output],
-    ).then(
-        process_url,
-        inputs=[
-            auth,
-            url_input,
-            to_formats,
-            image_export_mode,
-            pipeline,
-            ocr,
-            force_ocr,
-            ocr_engine,
-            ocr_lang,
-            pdf_backend,
-            table_mode,
-            abort_on_error,
-            return_as_file,
-            do_code_enrichment,
-            do_formula_enrichment,
-            do_picture_classification,
-            do_picture_description,
-        ],
-        outputs=[
-            task_id_rendered,
-        ],
-    ).then(
-        set_outputs_visibility_process,
-        inputs=[return_as_file],
-        outputs=[content_output, file_output],
-    ).then(
-        wait_task_finish,
-        inputs=[auth, task_id_rendered, return_as_file],
-        outputs=[
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-            download_file_btn,
-        ],
-    )
-
-    url_reset_btn.click(
-        clear_outputs,
-        inputs=None,
-        outputs=[
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-        ],
-    ).then(set_options_visibility, inputs=[true_bool], outputs=[options]).then(
-        set_outputs_visibility_direct,
-        inputs=[false_bool, false_bool],
-        outputs=[content_output, file_output],
-    ).then(set_task_id_visibility, inputs=[false_bool], outputs=[task_id_output]).then(
-        clear_url_input, inputs=None, outputs=[url_input]
-    )
-
-    # File processing
-    file_process_btn.click(
-        set_options_visibility, inputs=[false_bool], outputs=[options]
-    ).then(
-        set_download_button_label, inputs=[processing_text], outputs=[download_file_btn]
-    ).then(
-        clear_outputs,
-        inputs=None,
-        outputs=[
-            task_id_rendered,
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-        ],
-    ).then(
-        set_task_id_visibility,
-        inputs=[true_bool],
-        outputs=[task_id_output],
-    ).then(
-        process_file,
-        inputs=[
-            auth,
-            file_input,
-            to_formats,
-            image_export_mode,
-            pipeline,
-            ocr,
-            force_ocr,
-            ocr_engine,
-            ocr_lang,
-            pdf_backend,
-            table_mode,
-            abort_on_error,
-            return_as_file,
-            do_code_enrichment,
-            do_formula_enrichment,
-            do_picture_classification,
-            do_picture_description,
-        ],
-        outputs=[
-            task_id_rendered,
-        ],
-    ).then(
-        set_outputs_visibility_process,
-        inputs=[return_as_file],
-        outputs=[content_output, file_output],
-    ).then(
-        wait_task_finish,
-        inputs=[auth, task_id_rendered, return_as_file],
-        outputs=[
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-            download_file_btn,
-        ],
-    )
-
-    file_reset_btn.click(
-        clear_outputs,
-        inputs=None,
-        outputs=[
-            output_markdown,
-            output_markdown_rendered,
-            output_json,
-            output_json_rendered,
-            output_html,
-            output_html_rendered,
-            output_text,
-            output_doctags,
-        ],
-    ).then(set_options_visibility, inputs=[true_bool], outputs=[options]).then(
-        set_outputs_visibility_direct,
-        inputs=[false_bool, false_bool],
-        outputs=[content_output, file_output],
-    ).then(set_task_id_visibility, inputs=[false_bool], outputs=[task_id_output]).then(
-        clear_file_input, inputs=None, outputs=[file_input]
-    )
--- a/docling_serve/helper_functions.py
+++ b/docling_serve/helper_functions.py
@@ -1,11 +1,25 @@
+import importlib.metadata
 import inspect
 import json
+import platform
 import re
+import sys
 from typing import Union, get_args, get_origin

 from fastapi import Depends, Form
 from pydantic import BaseModel, TypeAdapter

+DOCLING_VERSIONS = {
+    "docling-serve": importlib.metadata.version("docling-serve"),
+    "docling-jobkit": importlib.metadata.version("docling-jobkit"),
+    "docling": importlib.metadata.version("docling"),
+    "docling-core": importlib.metadata.version("docling-core"),
+    "docling-ibm-models": importlib.metadata.version("docling-ibm-models"),
+    "docling-parse": importlib.metadata.version("docling-parse"),
+    "python": f"{sys.implementation.cache_tag} ({platform.python_version()})",
+    "plaform": platform.platform(),
+}
+

 def is_pydantic_model(type_):
    try:
--- a/docling_serve/orchestrator_factory.py
+++ b/docling_serve/orchestrator_factory.py
@@ -1,10 +1,267 @@
+import json
+import logging
 from functools import lru_cache
+from typing import Any, Optional

-from docling_jobkit.orchestrators.base_orchestrator import BaseOrchestrator
+import redis.asyncio as redis
+
+from docling_jobkit.datamodel.task import Task
+from docling_jobkit.datamodel.task_meta import TaskStatus
+from docling_jobkit.orchestrators.base_orchestrator import (
+    BaseOrchestrator,
+    TaskNotFoundError,
+)

 from docling_serve.settings import AsyncEngine, docling_serve_settings
 from docling_serve.storage import get_scratch

+_log = logging.getLogger(__name__)
+
+
+class RedisTaskStatusMixin:
+    tasks: dict[str, Task]
+    _task_result_keys: dict[str, str]
+    config: Any
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.redis_prefix = "docling:tasks:"
+        self._redis_pool = redis.ConnectionPool.from_url(
+            self.config.redis_url,
+            max_connections=10,
+            socket_timeout=2.0,
+        )
+
+    async def task_status(self, task_id: str, wait: float = 0.0) -> Task:
+        """
+        Get task status by checking Redis first, then falling back to RQ verification.
+
+        When Redis shows 'pending' but RQ shows 'success', we update Redis
+        and return the RQ status for cross-instance consistency.
+        """
+        _log.info(f"Task {task_id} status check")
+
+        # Always check RQ directly first - this is the most reliable source
+        rq_task = await self._get_task_from_rq_direct(task_id)
+        if rq_task:
+            _log.info(f"Task {task_id} in RQ: {rq_task.task_status}")
+
+            # Update memory registry
+            self.tasks[task_id] = rq_task
+
+            # Store/update in Redis for other instances
+            await self._store_task_in_redis(rq_task)
+            return rq_task
+
+        # If not in RQ, check Redis (maybe it's cached from another instance)
+        task = await self._get_task_from_redis(task_id)
+        if task:
+            _log.info(f"Task {task_id} in Redis: {task.task_status}")
+
+            # CRITICAL FIX: Check if Redis status might be stale
+            # STARTED tasks might have completed since they were cached
+            if task.task_status in [TaskStatus.PENDING, TaskStatus.STARTED]:
+                _log.debug(f"Task {task_id} verifying stale status")
+
+                # Try to get fresh status from RQ
+                fresh_rq_task = await self._get_task_from_rq_direct(task_id)
+                if fresh_rq_task and fresh_rq_task.task_status != task.task_status:
+                    _log.info(
+                        f"Task {task_id} status updated: {fresh_rq_task.task_status}"
+                    )
+
+                    # Update memory and Redis with fresh status
+                    self.tasks[task_id] = fresh_rq_task
+                    await self._store_task_in_redis(fresh_rq_task)
+                    return fresh_rq_task
+                else:
+                    _log.debug(f"Task {task_id} status consistent")
+
+            return task
+
+        # Fall back to parent implementation
+        try:
+            parent_task = await super().task_status(task_id, wait)  # type: ignore[misc]
+            _log.debug(f"Task {task_id} from parent: {parent_task.task_status}")
+
+            # Store in Redis for other instances to find
+            await self._store_task_in_redis(parent_task)
+            return parent_task
+        except TaskNotFoundError:
+            _log.warning(f"Task {task_id} not found")
+            raise
+
+    async def _get_task_from_redis(self, task_id: str) -> Optional[Task]:
+        try:
+            async with redis.Redis(connection_pool=self._redis_pool) as r:
+                task_data = await r.get(f"{self.redis_prefix}{task_id}:metadata")
+                if not task_data:
+                    return None
+
+                data: dict[str, Any] = json.loads(task_data)
+                meta = data.get("processing_meta") or {}
+                meta.setdefault("num_docs", 0)
+                meta.setdefault("num_processed", 0)
+                meta.setdefault("num_succeeded", 0)
+                meta.setdefault("num_failed", 0)
+
+                return Task(
+                    task_id=data["task_id"],
+                    task_type=data["task_type"],
+                    task_status=TaskStatus(data["task_status"]),
+                    processing_meta=meta,
+                )
+        except Exception as e:
+            _log.error(f"Redis get task {task_id}: {e}")
+            return None
+
+    async def _get_task_from_rq_direct(self, task_id: str) -> Optional[Task]:
+        try:
+            _log.debug(f"Checking RQ for task {task_id}")
+
+            temp_task = Task(
+                task_id=task_id,
+                task_type="convert",
+                task_status=TaskStatus.PENDING,
+                processing_meta={
+                    "num_docs": 0,
+                    "num_processed": 0,
+                    "num_succeeded": 0,
+                    "num_failed": 0,
+                },
+            )
+
+            original_task = self.tasks.get(task_id)
+            self.tasks[task_id] = temp_task
+
+            try:
+                await super()._update_task_from_rq(task_id)  # type: ignore[misc]
+
+                updated_task = self.tasks.get(task_id)
+                if updated_task and updated_task.task_status != TaskStatus.PENDING:
+                    _log.debug(f"RQ task {task_id}: {updated_task.task_status}")
+
+                    # Store result key if available
+                    if task_id in self._task_result_keys:
+                        try:
+                            async with redis.Redis(
+                                connection_pool=self._redis_pool
+                            ) as r:
+                                await r.set(
+                                    f"{self.redis_prefix}{task_id}:result_key",
+                                    self._task_result_keys[task_id],
+                                    ex=86400,
+                                )
+                                _log.debug(f"Stored result key for {task_id}")
+                        except Exception as e:
+                            _log.error(f"Store result key {task_id}: {e}")
+
+                    return updated_task
+                return None
+
+            finally:
+                # Restore original task state
+                if original_task:
+                    self.tasks[task_id] = original_task
+                elif task_id in self.tasks and self.tasks[task_id] == temp_task:
+                    # Only remove if it's still our temp task
+                    del self.tasks[task_id]
+
+        except Exception as e:
+            _log.error(f"RQ check {task_id}: {e}")
+            return None
+
+    async def get_raw_task(self, task_id: str) -> Task:
+        if task_id in self.tasks:
+            return self.tasks[task_id]
+
+        task = await self._get_task_from_redis(task_id)
+        if task:
+            self.tasks[task_id] = task
+            return task
+
+        try:
+            parent_task = await super().get_raw_task(task_id)  # type: ignore[misc]
+            await self._store_task_in_redis(parent_task)
+            return parent_task
+        except TaskNotFoundError:
+            raise
+
+    async def _store_task_in_redis(self, task: Task) -> None:
+        try:
+            meta: Any = task.processing_meta
+            if hasattr(meta, "model_dump"):
+                meta = meta.model_dump()
+            elif not isinstance(meta, dict):
+                meta = {
+                    "num_docs": 0,
+                    "num_processed": 0,
+                    "num_succeeded": 0,
+                    "num_failed": 0,
+                }
+
+            data: dict[str, Any] = {
+                "task_id": task.task_id,
+                "task_type": task.task_type.value
+                if hasattr(task.task_type, "value")
+                else str(task.task_type),
+                "task_status": task.task_status.value,
+                "processing_meta": meta,
+            }
+            async with redis.Redis(connection_pool=self._redis_pool) as r:
+                await r.set(
+                    f"{self.redis_prefix}{task.task_id}:metadata",
+                    json.dumps(data),
+                    ex=86400,
+                )
+        except Exception as e:
+            _log.error(f"Store task {task.task_id}: {e}")
+
+    async def enqueue(self, **kwargs):  # type: ignore[override]
+        task = await super().enqueue(**kwargs)  # type: ignore[misc]
+        await self._store_task_in_redis(task)
+        return task
+
+    async def task_result(self, task_id: str):  # type: ignore[override]
+        result = await super().task_result(task_id)  # type: ignore[misc]
+        if result is not None:
+            return result
+
+        try:
+            async with redis.Redis(connection_pool=self._redis_pool) as r:
+                result_key = await r.get(f"{self.redis_prefix}{task_id}:result_key")
+                if result_key:
+                    self._task_result_keys[task_id] = result_key.decode("utf-8")
+                    return await super().task_result(task_id)  # type: ignore[misc]
+        except Exception as e:
+            _log.error(f"Redis result key {task_id}: {e}")
+
+        return None
+
+    async def _update_task_from_rq(self, task_id: str) -> None:
+        original_status = (
+            self.tasks[task_id].task_status if task_id in self.tasks else None
+        )
+
+        await super()._update_task_from_rq(task_id)  # type: ignore[misc]
+
+        if task_id in self.tasks:
+            new_status = self.tasks[task_id].task_status
+            if original_status != new_status:
+                _log.debug(f"Task {task_id} status: {original_status} -> {new_status}")
+                await self._store_task_in_redis(self.tasks[task_id])
+
+        if task_id in self._task_result_keys:
+            try:
+                async with redis.Redis(connection_pool=self._redis_pool) as r:
+                    await r.set(
+                        f"{self.redis_prefix}{task_id}:result_key",
+                        self._task_result_keys[task_id],
+                        ex=86400,
+                    )
+            except Exception as e:
+                _log.error(f"Store result key {task_id}: {e}")
+

@lru_cache
 def get_async_orchestrator() -> BaseOrchestrator:
@@ -31,16 +288,25 @@ def get_async_orchestrator() -> BaseOrchestrator:
            allow_external_plugins=docling_serve_settings.allow_external_plugins,
            max_num_pages=docling_serve_settings.max_num_pages,
            max_file_size=docling_serve_settings.max_file_size,
+            queue_max_size=docling_serve_settings.queue_max_size,
+            ocr_batch_size=docling_serve_settings.ocr_batch_size,
+            layout_batch_size=docling_serve_settings.layout_batch_size,
+            table_batch_size=docling_serve_settings.table_batch_size,
+            batch_polling_interval_seconds=docling_serve_settings.batch_polling_interval_seconds,
        )
        cm = DoclingConverterManager(config=cm_config)

        return LocalOrchestrator(config=local_config, converter_manager=cm)
+
    elif docling_serve_settings.eng_kind == AsyncEngine.RQ:
        from docling_jobkit.orchestrators.rq.orchestrator import (
            RQOrchestrator,
            RQOrchestratorConfig,
        )

+        class RedisAwareRQOrchestrator(RedisTaskStatusMixin, RQOrchestrator):  # type: ignore[misc]
+            pass
+
        rq_config = RQOrchestratorConfig(
            redis_url=docling_serve_settings.eng_rq_redis_url,
            results_prefix=docling_serve_settings.eng_rq_results_prefix,
@@ -48,7 +314,8 @@ def get_async_orchestrator() -> BaseOrchestrator:
            scratch_dir=get_scratch(),
        )

-        return RQOrchestrator(config=rq_config)
+        return RedisAwareRQOrchestrator(config=rq_config)
+
    elif docling_serve_settings.eng_kind == AsyncEngine.KFP:
        from docling_jobkit.orchestrators.kfp.orchestrator import (
            KfpOrchestrator,
--- a/docling_serve/settings.py
+++ b/docling_serve/settings.py
@@ -50,6 +50,7 @@ class DoclingServeSettings(BaseSettings):
    options_cache_size: int = 2
    enable_remote_services: bool = False
    allow_external_plugins: bool = False
+    show_version_info: bool = True

    api_key: str = ""

@@ -57,6 +58,14 @@ class DoclingServeSettings(BaseSettings):
    max_num_pages: int = sys.maxsize
    max_file_size: int = sys.maxsize

+    # Threading pipeline
+    queue_max_size: Optional[int] = None
+    ocr_batch_size: Optional[int] = None
+    layout_batch_size: Optional[int] = None
+    table_batch_size: Optional[int] = None
+    batch_polling_interval_seconds: Optional[float] = None
+
+    sync_poll_interval: int = 2  # seconds
    max_sync_wait: int = 120  # 2 minutes

    cors_origins: list[str] = ["*"]
--- a/docling_serve/ui/init.py
+++ b/docling_serve/ui/init.py
--- a/docling_serve/ui/app.py
+++ b/docling_serve/ui/app.py
@@ -0,0 +1,278 @@
+import io
+import logging
+from pathlib import Path
+from typing import Annotated
+
+from fastapi import (
+    BackgroundTasks,
+    Depends,
+    FastAPI,
+    Form,
+    HTTPException,
+    Request,
+    UploadFile,
+    status,
+)
+from fastapi.responses import HTMLResponse, RedirectResponse, Response
+from fastapi.staticfiles import StaticFiles
+from pydantic import AnyHttpUrl
+from pyjsx import auto_setup  # type: ignore
+from starlette.exceptions import HTTPException as StarletteHTTPException
+
+from docling.datamodel.base_models import OutputFormat
+from docling_core.types.doc.document import (
+    FloatingItem,
+    PageItem,
+    RefItem,
+)
+from docling_jobkit.orchestrators.base_orchestrator import (
+    BaseOrchestrator,
+)
+
+from docling_serve.auth import APIKeyCookieAuth, AuthenticationResult
+from docling_serve.datamodel.convert import ConvertDocumentsRequestOptions
+from docling_serve.datamodel.requests import ConvertDocumentsRequest, HttpSourceRequest
+from docling_serve.helper_functions import FormDepends
+from docling_serve.orchestrator_factory import get_async_orchestrator
+from docling_serve.settings import docling_serve_settings
+
+from .convert import ConvertPage  # type: ignore
+from .pages import AuthPage, StatusPage, TaskPage, TasksPage  # type: ignore
+
+# Initialize JSX.
+auto_setup
+
+_log = logging.getLogger(__name__)
+
+
+# TODO: Isolate passed functions into a controller?
+def create_ui_app(process_file, process_url, task_result, task_status_poll) -> FastAPI:  # noqa: C901
+    ui_app = FastAPI()
+    require_auth = APIKeyCookieAuth(docling_serve_settings.api_key)
+
+    # Static files.
+    ui_app.mount(
+        "/static",
+        StaticFiles(directory=Path(__file__).parent.absolute() / "static"),
+        name="static",
+    )
+
+    # Convert page.
+    @ui_app.get("/")
+    async def get_root():
+        return RedirectResponse(url="convert")
+
+    @ui_app.get("/convert", response_class=HTMLResponse)
+    async def get_convert(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+    ):
+        return str(ConvertPage())
+
+    @ui_app.post("/convert", response_class=HTMLResponse)
+    async def post_convert(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        background_tasks: BackgroundTasks,
+        options: Annotated[
+            ConvertDocumentsRequestOptions, FormDepends(ConvertDocumentsRequestOptions)
+        ],
+        files: Annotated[list[UploadFile], Form()],
+        url: Annotated[str, Form()],
+        page_min: Annotated[str, Form()],
+        page_max: Annotated[str, Form()],
+    ):
+        # Refined model options and behavior.
+        if len(page_min) > 0:
+            options.page_range = (int(page_min), options.page_range[1])
+        if len(page_max) > 0:
+            options.page_range = (options.page_range[0], int(page_max))
+
+        options.ocr_lang = [
+            sub_lang.strip()
+            for lang in options.ocr_lang or []
+            for sub_lang in lang.split(",")
+            if len(sub_lang.strip()) > 0
+        ]
+
+        files = [f for f in files if f.size]
+        if len(files) > 0:
+            # Directly uploaded documents.
+            response = await process_file(
+                auth=auth,
+                orchestrator=orchestrator,
+                background_tasks=background_tasks,
+                files=files,
+                options=options,
+            )
+        elif len(url.strip()) > 0:
+            # URLs of documents.
+            source = HttpSourceRequest(url=AnyHttpUrl(url))
+            request = ConvertDocumentsRequest(options=options, sources=[source])
+
+            response = await process_url(
+                auth=auth,
+                orchestrator=orchestrator,
+                conversion_request=request,
+            )
+        else:
+            validation = {
+                "files": "Upload files or enter a URL",
+                "url": "Enter a URL or upload files",
+            }
+            return str(ConvertPage(options=options, validation=validation))
+
+        return RedirectResponse(f"tasks/{response.task_id}/", status.HTTP_303_SEE_OTHER)
+
+    # Task overview page.
+    @ui_app.get("/tasks/", response_class=HTMLResponse)
+    async def get_tasks(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+    ):
+        tasks = sorted(orchestrator.tasks.values(), key=lambda t: t.created_at)
+
+        return str(TasksPage(tasks))
+
+    # Task specific page.
+    @ui_app.get("/tasks/{task_id}/", response_class=HTMLResponse)
+    async def get_task(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        background_tasks: BackgroundTasks,
+        task_id: str,
+    ):
+        poll = await task_status_poll(auth, orchestrator, task_id)
+
+        result = None
+        if poll.task_status in ["success", "failure"]:
+            try:
+                result = await task_result(
+                    auth, orchestrator, background_tasks, task_id
+                )
+            except Exception as ex:
+                logging.error(ex)
+
+        return str(TaskPage(poll, result))
+
+    # Poll task via HTTP status.
+    @ui_app.get("/tasks/{task_id}/poll", response_class=Response)
+    async def poll_task(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        task_id: str,
+    ):
+        poll = await task_status_poll(auth, orchestrator, task_id)
+        return Response(
+            status_code=status.HTTP_202_ACCEPTED
+            if poll.task_status == "started"
+            else status.HTTP_200_OK
+        )
+
+    # Download the contents of zipped documents.
+    @ui_app.get("/tasks/{task_id}/documents.zip")
+    async def get_task_zip(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        background_tasks: BackgroundTasks,
+        task_id: str,
+    ):
+        return await task_result(auth, orchestrator, background_tasks, task_id)
+
+    # Get the output of a task, as a converted document in a specific format.
+    @ui_app.get("/tasks/{task_id}/document.{format}")
+    async def get_task_document_format(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        background_tasks: BackgroundTasks,
+        task_id: str,
+        format: str,
+    ):
+        if format not in [f.value for f in OutputFormat]:
+            raise HTTPException(status.HTTP_404_NOT_FOUND, "Output format not found.")
+        else:
+            response = await task_result(auth, orchestrator, background_tasks, task_id)
+
+        # TODO: Make this compatible with base_models FormatToMimeType?
+        mimes = {
+            "html": "text/html",
+            "md": "text/markdown",
+            "json": "application/json",
+        }
+
+        content = (
+            response.document.json_content.export_to_dict()
+            if format == OutputFormat.JSON
+            else response.document.dict()[f"{format}_content"]
+        )
+
+        return Response(
+            content=str(content),
+            media_type=mimes.get(format, "text/plain"),
+        )
+
+    @ui_app.get("/tasks/{task_id}/document/{cref:path}")
+    async def get_task_document_item(
+        request: Request,
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+        orchestrator: Annotated[BaseOrchestrator, Depends(get_async_orchestrator)],
+        background_tasks: BackgroundTasks,
+        task_id: str,
+        cref: str,
+    ):
+        response = await task_result(auth, orchestrator, background_tasks, task_id)
+        doc = response.document.json_content
+        item = RefItem(cref=f"#/{cref}").resolve(doc)  # type: ignore
+
+        if "image/*" in (request.headers.get("Accept") or "") and isinstance(
+            item, FloatingItem | PageItem
+        ):
+            content = io.BytesIO()
+
+            if (
+                isinstance(item, PageItem)
+                and (img_ref := item.image)
+                and img_ref.pil_image
+            ):
+                img_ref.pil_image.save(content, format="PNG")
+            elif isinstance(item, FloatingItem) and (img := item.get_image(doc)):
+                img.save(content, format="PNG")
+
+            return Response(content=content.getvalue(), media_type="image/png")
+        else:
+            return item
+
+    # Page not found; catch all.
+    @ui_app.api_route("/{path_name:path}")
+    def no_page(
+        auth: Annotated[AuthenticationResult, Depends(require_auth)],
+    ):
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Page not found.")
+
+    # Exception and auth pages.
+    @ui_app.exception_handler(StarletteHTTPException)
+    @ui_app.exception_handler(Exception)
+    async def exception_page(request: Request, ex: Exception):
+        if not isinstance(ex, StarletteHTTPException):
+            # Internal error.
+            ex = HTTPException(status.HTTP_500_INTERNAL_SERVER_ERROR)
+
+        if request.method == "POST":
+            # Authorization required -> API key dialog.
+            form = await request.form()
+            form_api_key = form.get("api_key")
+            if isinstance(form_api_key, str):
+                response = RedirectResponse(request.url, status.HTTP_303_SEE_OTHER)
+                require_auth._set_api_key(response, form_api_key)
+                return response
+
+        if ex.status_code == status.HTTP_401_UNAUTHORIZED:
+            return HTMLResponse(str(AuthPage()), status.HTTP_401_UNAUTHORIZED)
+
+        # HTTP exception page; avoid referer loop.
+        referer = request.headers.get("Referer")
+        if referer == request.url:
+            referer = None
+
+        return HTMLResponse(str(StatusPage(ex, referer)), ex.status_code)
+
+    return ui_app
--- a/docling_serve/ui/convert.px
+++ b/docling_serve/ui/convert.px
@@ -0,0 +1,251 @@
+import json
+import sys
+
+from pyjsx import jsx, JSX
+
+from docling.datamodel.base_models import FormatToExtensions, OutputFormat
+from docling.datamodel.pipeline_options import PdfBackend, ProcessingPipeline, TableFormerMode
+from docling_core.types.doc import ImageRefMode
+from docling_serve.datamodel.convert import ConvertDocumentsRequestOptions, ocr_engines_enum
+
+from .forms import EnumCheckboxes, EnumRadios, EnumSelect, ocr_engine_languages, ValidatedInput
+from .pages import Header, Page
+
+base_convert_options = ConvertDocumentsRequestOptions()
+base_convert_options.to_formats.append(OutputFormat.JSON)
+
+
+def ConvertPage(
+  options: ConvertDocumentsRequestOptions = base_convert_options,
+  validation: None | dict[str, str] = None
+) -> JSX:
+    file_accept = ",".join([f".{ext}" for exts in FormatToExtensions.values() for ext in exts])
+
+    return (
+        <Page title="Convert">
+            <main class="container">
+                <Header />
+
+                <form class="convert" method="post" enctype="multipart/form-data">
+                    <legend>
+                        <b>Documents</b>
+                    </legend>
+                    <fieldset class="grid">
+                        <ValidatedInput
+                            name="files"
+                            type="file"
+                            multiple
+                            accept={file_accept}
+                            validation={validation}
+                        />
+                        <ValidatedInput
+                            name="url"
+                            placeholder="or enter a URL: https://arxiv.org/pdf/2501.17887"
+                            validation={validation}
+                        />
+                    </fieldset>
+
+                    <fieldset class="grid">
+                        <EnumSelect
+                            enum={ProcessingPipeline}
+                            selected={options.pipeline}
+                            name="pipeline"
+                            title="Pipeline"
+                        />
+                        <EnumSelect
+                            enum={PdfBackend}
+                            selected={options.pdf_backend}
+                            name="pdf_backend"
+                            title="PDF Backend"
+                        />
+
+                        <div>
+                            <label>Pages</label>
+                            <div role="group">
+                                <input
+                                    type="number"
+                                    name="page_min"
+                                    min={1}
+                                    step={1}
+                                    placeholder="1"
+                                    value={None if options.page_range[0] <= 1 else options.page_range[0]}
+                                />
+                                <input
+                                    type="number"
+                                    name="page_max"
+                                    min={1}
+                                    step={1}
+                                    placeholder="max."
+                                    value={None if options.page_range[1] >= sys.maxsize else options.page_range[1]}
+                                />
+                            </div>
+                        </div>
+
+                        <div>
+                            <label>Timeout<small>in seconds</small></label>
+                            <input
+                                type="number"
+                                name="document_timeout"
+                                min={1}
+                                step={1}
+                                value={int(options.document_timeout)}
+                            />
+                        </div>
+                    </fieldset>
+
+                    <div class="grid">
+                        <EnumCheckboxes
+                            enum={OutputFormat}
+                            selected={options.to_formats}
+                            name="to_formats"
+                            title={<b>Output</b>}
+                        />
+
+                        <div>
+                            <fieldset>
+                                <label>
+                                    <input
+                                        type="checkbox"
+                                        name="do_ocr"
+                                        checked={options.do_ocr}
+                                    />
+                                    <b>OCR</b>
+                                </label>
+                                <label display-when="do_ocr">
+                                    <input
+                                        type="checkbox"
+                                        name="do_code_enrichment"
+                                        checked={options.do_code_enrichment}
+                                    />
+                                    Code
+                                </label>
+                                <label display-when="do_ocr">
+                                    <input
+                                        type="checkbox"
+                                        name="do_formula_enrichment"
+                                        checked={options.do_formula_enrichment}
+                                    />
+                                    Formulas
+                                </label>
+                            </fieldset>
+
+                            <EnumSelect
+                                display-when="do_ocr"
+                                enum={ocr_engines_enum}
+                                selected={options.ocr_engine}
+                                name="ocr_engine"
+                                title="Engine"
+                            />
+
+                            <label display-when="do_ocr">Language</label>
+                            <input
+                                display-when="do_ocr"
+                                name="ocr_lang"
+                                dep-on="ocr_engine"
+                                dep-values={json.dumps(ocr_engine_languages)}
+                                pattern="[\w+]*[,\w+]*"
+                                title="A comma separated list of language codes, of which the format depends on the selected engine."
+                            />
+
+                            <label display-when="do_ocr">
+                                <input
+                                    type="checkbox"
+                                    name="force_ocr"
+                                    checked={options.force_ocr}
+                                />
+                                Force
+                            </label>
+                        </div>
+
+                        <div>
+                            <fieldset>
+                                <label>
+                                    <input
+                                        type="checkbox"
+                                        name="include_images"
+                                        checked={options.include_images}
+                                    />
+                                    <b>Images</b>
+                                </label>
+                                <label display-when="include_images">
+                                    <input
+                                        type="checkbox"
+                                        name="do_picture_classification"
+                                        checked={options.do_picture_classification}
+                                    />
+                                    Classification
+                                </label>
+                                <label display-when="include_images">
+                                    <input
+                                        type="checkbox"
+                                        name="do_picture_description"
+                                        checked={options.do_picture_description}
+                                    />
+                                    Description
+                                </label>
+                                <label display-when="include_images,do_picture_description">Area threshold</label>
+                                <input
+                                    display-when="include_images,do_picture_description"
+                                    type="number"
+                                    name="picture_description_area_threshold"
+                                    min={0}
+                                    max={1}
+                                    step={0.01}
+                                    value={options.picture_description_area_threshold}
+                                />
+                            </fieldset>
+
+                            <EnumSelect
+                                display-when="include_images"
+                                enum={ImageRefMode}
+                                selected={options.image_export_mode}
+                                name="image_export_mode"
+                                title="Export"
+                            />
+                            <label display-when="include_images">Scale</label>
+                            <input
+                                display-when="include_images"
+                                type="number"
+                                name="images_scale"
+                                min={0}
+                                step={0.1}
+                                value={options.images_scale}
+                            />
+                        </div>
+
+                        <div>
+                            <fieldset>
+                                <label>
+                                    <input
+                                        type="checkbox"
+                                        name="do_table_structure"
+                                        checked={options.do_table_structure}
+                                    />
+                                    <b>Tables</b>
+                                </label>
+                                <label display-when="do_table_structure">
+                                    <input
+                                        type="checkbox"
+                                        name="table_cell_matching"
+                                        checked={options.table_cell_matching}
+                                    />
+                                    Cell matching
+                                </label>
+                            </fieldset>
+                            <EnumSelect
+                                display-when="do_table_structure"
+                                enum={TableFormerMode}
+                                selected={options.table_mode}
+                                name="table_mode"
+                                title="Mode"
+                            />
+                        </div>
+                    </div>
+
+                    <div class="sticky-footer">
+                        <input type="submit" value="Convert" />
+                    </div>
+                </form>
+            </main>
+        </Page>
+    )
--- a/docling_serve/ui/forms.px
+++ b/docling_serve/ui/forms.px
@@ -0,0 +1,127 @@
+from enum import Enum
+from typing import Type
+
+from pyjsx import jsx, JSX
+
+from docling.datamodel.pipeline_options import OcrOptions
+
+from docling_serve.datamodel.convert import ConvertDocumentsRequestOptions
+
+
+ocr_engine_languages = {
+    SubOptions.kind: ",".join(SubOptions().lang)
+    for SubOptions in OcrOptions.__subclasses__()
+}
+
+
+def _format_label(label: str) -> str:
+    return label.replace("_", " ").lower()
+
+
+def option_example(field_name: str) -> str | None:
+    field = ConvertDocumentsRequestOptions.model_fields[field_name]
+    return (field.examples or [])[0]
+
+
+def ValidatedInput(validation: None | dict[str, str], name: str, **kwargs) -> JSX:
+    if validation:
+        invalid = "true" if name in validation else "false"
+        content = [<input name={name} aria-invalid={invalid} {...kwargs} />]
+
+        if name in validation:
+            content.append(<small>{validation[name]}</small>)
+
+        return <div>{content}</div>
+    else:
+        return <input name={name} {...kwargs} />
+
+
+def EnumCheckboxes(
+    children,
+    enum: Type[Enum],
+    selected: list[Enum],
+    name: str,
+    title: JSX = None,
+    **kwargs
+) -> JSX:
+    return (
+        <fieldset {...kwargs}>
+            {
+                <legend>{title}</legend>
+                if title
+                else None
+            }
+
+            {[
+                <label>
+                    <input
+                        type="checkbox"
+                        name={name}
+                        value={e.value}
+                        checked={e.value in selected}
+                    />
+                    {_format_label(e.name)}
+                </label>
+                for e in enum
+            ]}
+        </fieldset>
+    )
+
+
+def EnumRadios(
+    children,
+    enum: Type[Enum],
+    selected: Enum,
+    name: str,
+    title: JSX = None,
+    **kwargs
+) -> JSX:
+    return (
+        <fieldset {...kwargs}>
+            {
+                <legend>{title}</legend>
+                if title
+                else None
+            }
+
+            {[
+                <label>
+                    <input
+                        type="radio"
+                        name={name}
+                        value={e.value}
+                        checked={e.value == selected}
+                    />
+                    {_format_label(e.name)}
+                </label>
+                for e in enum
+            ]}
+        </fieldset>
+    )
+
+
+def EnumSelect(
+    children,
+    enum: Type[Enum],
+    selected: Enum,
+    name: str,
+    title: JSX = None,
+    **kwargs
+) -> JSX:
+    return (
+        <div {...kwargs}>
+            {
+                <label>{title}</label>
+                if title
+                else None
+            }
+            <select name={name}>
+                {[
+                    <option value={e.value} selected={e.value == selected}>
+                        {_format_label(e.name)}
+                    </option>
+                    for e in enum
+                ]}
+            </select>
+        </div>
+    )
--- a/docling_serve/ui/pages.px
+++ b/docling_serve/ui/pages.px
@@ -0,0 +1,220 @@
+from importlib import metadata
+
+from fastapi import FastAPI, HTTPException, Response
+from pyjsx import jsx, JSX
+
+from docling.datamodel.base_models import OutputFormat
+from docling.datamodel.pipeline_options import PdfBackend, ProcessingPipeline, TableFormerMode
+from docling_jobkit.datamodel.task import Task
+from docling_serve.datamodel.responses import ConvertDocumentResponse
+
+from .preview import DocPreview
+
+
+def Header(children, classname: str = "") -> JSX:
+    return (
+        <header class={classname}>
+            <span class="title">
+                D<img src="/ui/static/logo.svg" />CLING SERVE
+            </span>
+
+            <span class="version" title="Docling version">
+                {metadata.version('docling')}
+            </span>
+
+            <nav>
+                <ul>
+                    <li><a href="/ui/convert">Convert</a></li>
+                    <li><a href="/ui/tasks/">Tasks</a></li>
+                </ul>
+            </nav>
+        </header>
+    )
+
+
+def Page(children, title: str, poll: bool = False) -> JSX:
+    return (
+        <html lang="en" id="root">
+            <head>
+                <title>{title}</title>
+
+                <meta charset="utf-8" />
+                <meta name="viewport" content="width=device-width, initial-scale=1" />
+
+                <link rel="stylesheet" href="/ui/static/style.css" />
+                <script src="/ui/static/main.js" />
+            </head>
+
+            <body onload={'setInterval(async () => { if ((await fetch("poll")).status == 200) location.reload(); }, 3000)' if poll else None}>
+                {children}
+            </body>
+        </html>
+    )
+
+
+def AuthPage():
+    return (
+        <Page title="Authenticate">
+            <form method="post">
+                <dialog open>
+                    <article>
+                        <header>
+                            <h4>Authenticate</h4>
+                        </header>
+                        <input
+                            type="password"
+                            name="api_key"
+                            placeholder="Enter API key"
+                            required autofocus
+                        />
+                        <footer>
+                            <input type="submit" value="Confirm" />
+                        </footer>
+                    </article>
+                </dialog>
+            </form>
+        </Page>
+    )
+
+
+def TasksPage(tasks: list[Task]) -> JSX:
+    return (
+        <Page title="Tasks">
+            <main class="container">
+                <Header />
+
+                {(
+                    <p>There are no active tasks. <a href="../convert">Convert</a> a document to create a new task.</p>
+                ) if len(tasks) == 0 else (
+                    <table>
+                        <thead>
+                            <tr>
+                                <th>Task</th>
+                                <th>Status</th>
+                                <th>ID</th>
+                                <th>Created</th>
+                            </tr>
+                        </thead>
+                        <tbody>
+                            {
+                                <tr>
+                                    <td>{task.task_type.name}</td>
+                                    <td>{task.task_status.name}</td>
+                                    <td>
+                                        <a href={f"{task.task_id}/"}>{task.task_id}</a>
+                                    </td>
+                                    <td>{task.created_at.strftime("%d-%m-%Y, %H:%M:%S")}</td>
+                                </tr>
+                                for task in tasks
+                            }
+                        </tbody>
+                    </table>
+                )}
+            </main>
+        </Page>
+    )
+
+
+def TaskPage(poll, task: ConvertDocumentResponse) -> JSX:
+    def PlainPage(children, poll = False) -> JSX:
+        return (
+            <Page title="Task" poll={poll}>
+                <main class="container">
+                    <Header classname={"loading" if poll else None} />
+                    {children}
+                </main>
+            </Page>
+        )
+
+    if isinstance(task, Response):
+        return (
+            <PlainPage>
+                <p>
+                    <ins>Converted multiple documents successfully</ins>
+                </p>
+                <a href="documents.zip">documents.zip</a>
+            </PlainPage>
+        )
+    else:
+        match poll.task_status:
+            case "success":
+                doc = task.document.dict()
+                doc_json = task.document.json_content
+
+                return (
+                    <Page title={task.document.filename}>
+                        <main class="preview">
+                            <Header />
+
+                            <div class="status">
+                                <div>
+                                    <span>Task</span>
+                                    <b>{poll.task_id}</b>
+                                </div>
+                                <div>
+                                    <span>converted</span>
+                                    <b>{task.document.filename}</b>
+                                </div>
+                                <div>
+                                    <span>in</span>
+                                    <b>{round(task.processing_time)} seconds</b>
+                                </div>
+                            </div>
+                            
+                            <div class="formats">
+                                {[
+                                    <a class="secondary" href={f"document.{f.value}"} target="_blank">
+                                        <button>{f.name}</button>
+                                    </a>
+                                    for f in OutputFormat
+                                    if doc.get(f"{f.value}_content")
+                                ]}
+                                <label class="configDarkImg">
+                                    <input type="checkbox" name="invert-images" persist="preview" />
+                                    Invert images
+                                </label>
+                            </div>
+
+                            {
+                                <DocPreview doc={doc_json} />
+                                if doc_json
+                                else (<p>No document preview because JSON is missing as an output format.</p>)
+                            }
+                        </main>
+                    </Page>
+                )
+            case "started":
+                return (
+                    <PlainPage poll>
+                        <p class="progress">Task <b>{poll.task_id}</b> is in progress...</p>
+                        <progress />
+                    </PlainPage>
+                )
+            case _:
+                return (
+                    <PlainPage>
+                        <p class="fail">
+                            Task <b>{poll.task_id}</b> failed.
+                        </p>
+                        <button onclick="history.back()">
+                            Go back
+                        </button>
+                    </PlainPage>
+                )
+
+
+def StatusPage(ex: HTTPException, referer: str | None) -> JSX:
+    return (
+        <Page title={ex.status_code}>
+            <main class="container">
+                <Header />
+                <h4>{ex.status_code}</h4>
+                <p>{ex.detail}</p>
+                <p>
+                    <a href={referer or ".."}>
+                        <button>Go back</button>
+                    </a>
+                </p>
+            </main>
+        </Page>
+    )
--- a/docling_serve/ui/preview.px
+++ b/docling_serve/ui/preview.px
@@ -0,0 +1,347 @@
+from collections import defaultdict
+from html import escape
+from typing import Type
+
+from docling_core.types.doc.document import (
+    BaseAnnotation,
+    CodeItem,
+    ContentLayer,
+    DescriptionAnnotation,
+    DoclingDocument,
+    DocItem,
+    FloatingItem,
+    Formatting,
+    FormulaItem,
+    GroupItem,
+    GroupLabel,
+    ListGroup,
+    ListItem,
+    NodeItem,
+    PictureClassificationData,
+    PictureItem,
+    ProvenanceItem,
+    RefItem,
+    Script,
+    SectionHeaderItem,
+    TableCell,
+    TableItem,
+    TextItem,
+    TitleItem
+)
+from pyjsx import jsx, JSX, JSXComponent
+
+from .svg import image, path, rect, text
+
+
+_node_components: dict[str, JSXComponent] = {}
+
+
+def component(*node_types: list[Type[BaseAnnotation | NodeItem]]):
+    def decorator(component):
+        for t in node_types:
+            _node_components[t.__name__] = component
+    return decorator
+
+
+def AnnotationComponent(children, annotation: BaseAnnotation):
+    Comp = _node_components.get(annotation.__class__.__name__)
+    element = Comp(annotation=annotation, children=[]) if Comp else (
+        <code>{escape(annotation.model_dump_json(indent=2))}</code>
+    )
+
+    element.props["class"] = element.props.get("class", "") + " annotation"
+    element.props["data-kind"] = annotation.kind
+
+    return element
+
+
+def NodeComponent(children, node: NodeItem | RefItem, doc: DoclingDocument):
+    # Specific component or fallback.
+    Comp = _node_components.get(node.__class__.__name__)
+    element = Comp(node=node, doc=doc, children=[]) if Comp else (
+        <span class="void"></span>
+    )
+
+    # Wrap item component with annotations, if any.
+    if isinstance(node, DocItem) and (anns := node.get_annotations()):
+        element = (
+            <div class="annotated">
+                {element}
+                {[<AnnotationComponent annotation={ann} /> for ann in anns]}
+            </div>
+        )
+
+    # Extend interaction and styling.
+    id = node.self_ref[2:]
+    element.props["id"] = id
+    element.props["onclick"] = "clickId(event)"
+
+    classes = ["item", node.content_layer.value]
+    element.props["class"] = f"{element.props.get("class", "")} {" ".join(classes)}"
+
+    return element
+
+
+def node_provs(node: NodeItem, doc: DoclingDocument) -> ProvenanceItem:
+    return node.prov if isinstance(node, DocItem) else [
+        p
+        for c in node.children
+        if isinstance(c.resolve(doc), DocItem)
+        for p in c.resolve(doc).prov
+    ]
+
+
+def DocPage(children, page_no: int, items: list[NodeItem], doc: DoclingDocument):
+    page = doc.pages[page_no]
+    exclusive_items = [
+        item
+        for item in items
+        if min([p.page_no for p in node_provs(item, doc)]) == page_no
+    ]
+
+    comps = []
+    for i in range(len(exclusive_items)):
+        item = exclusive_items[i]
+        id = item.self_ref[2:]
+        kind, *index = id.split("/")
+
+        parent_class = ""
+        if isinstance(item, GroupItem):
+            parent_class = "group"
+        else:
+            parent = item.parent.resolve(doc)
+            if isinstance(parent, GroupItem) and parent.label is not GroupLabel.UNSPECIFIED:
+                parent_class = "grouped"
+
+        comps.append(
+            <div class={f"item-markers {parent_class} {item.content_layer.value}"} data-id={id}>
+                <span>{"/".join(index)}</span>
+                <span>{item.label.replace("_", " ")}</span>
+                {
+                    <span>{item.content_layer.value.replace("_", " ")}</span>
+                    if item.content_layer is not ContentLayer.BODY
+                    else None
+                }
+                <a href={f"document/{id}"} target="_blank">{"{;}"}</a>
+            </div>
+        )
+        comps.append(<NodeComponent node={item} doc={doc} />)
+
+        pages = set([p.page_no for p in node_provs(item, doc)])
+        page_mark_class = "page-marker"
+        if i == 0 or len(pages) > 1:
+            page_mark_class += " border"
+        comps.append(<div class={page_mark_class}></div>)
+
+    
+    def ItemBox(children, item: DocItem, prov: ProvenanceItem):
+        item_id = item.self_ref[2:]
+        sub_items = [
+            (item_id, prov.bbox.to_top_left_origin(page.size.height))
+        ]
+
+        # Table cells.
+        if isinstance(item, TableItem):
+            for cell in item.data.table_cells:
+                sub_items.append(
+                    (f"{item_id}/{cell.start_col_offset_idx}/{cell.start_row_offset_idx}", cell.bbox)
+                )
+
+        return [
+            <rect
+                data-id={id}
+                x={bbox.l - 1}
+                y={bbox.t - 1}
+                width={bbox.width + 2}
+                height={bbox.height + 2}
+                vector-effect="non-scaling-stroke"
+                onclick="clickId(event)"
+            />
+            for id, bbox in sub_items
+        ]
+
+    # Span extra row to fill up excess space.
+    comps.append(
+        <svg
+            class="page-image"
+            style={{ "grid-row": f"span {len(exclusive_items) + 1}" }}
+            width="50vw"
+            viewBox={f"0 0 {page.size.width} {page.size.height}"}
+        >
+            <image
+                href={f"document/pages/{page_no}"}
+                width={page.size.width}
+                height={page.size.height}
+            />
+            {[
+                <ItemBox item={item} prov={prov} />
+                for item in items
+                if isinstance(item, DocItem)
+                for prov in item.prov
+                if prov.page_no == page_no
+            ]}
+
+            <text class="top-no" x={5} y={5}>{page_no}</text>
+            <text class="bottom-no" x={5} y={page.size.height - 5}>{page_no}</text>
+        </svg>
+    )
+
+    return <div class="page">{comps}</div>
+
+
+def DocPreview(children, doc: DoclingDocument):
+    page_items: dict[int, list[NodeItem]] = defaultdict(list)
+
+    for item, level in doc.iterate_items(
+        with_groups=True,
+        included_content_layers={*ContentLayer}
+    ):
+        if not isinstance(item, GroupItem) or item.label is not GroupLabel.UNSPECIFIED:
+            pages = set([p.page_no for p in node_provs(item, doc)])
+            for page in pages:
+                page_items[page].append(item)
+
+    return [
+        <DocPage page_no={page_no} items={page_items[page_no]} doc={doc} />
+        for page_no in sorted(page_items.keys())
+    ]
+
+
+def _text_classes(node: TextItem) -> str:
+    classes = [node.label]
+
+    if frmt := node.formatting:
+        formats = {
+            "bold": frmt.bold,
+            "italic": frmt.italic,
+            "underline": frmt.underline,
+            "strikethrough": frmt.strikethrough
+        }
+        classes.extend([cls for cls, active in formats.items() if active])
+        classes.append(frmt.script)
+
+    return " ".join(classes)
+
+
+@component(TextItem)
+def TextComponent(children, node: TextItem, doc: DoclingDocument):
+    return <p class={_text_classes(node)}>{escape(node.text)}</p>
+
+
+@component(TitleItem)
+def TitleComponent(children, node: TitleItem, doc: DoclingDocument):
+    return <h1 class={_text_classes(node)}>{escape(node.text)}</h1>
+
+
+@component(SectionHeaderItem)
+def SectionHeaderComponent(children, node: SectionHeaderItem, doc: DoclingDocument):
+    return <h4 class={_text_classes(node)}>{escape(node.text)}</h4>
+
+
+@component(ListItem)
+def ListComponent(children, node: ListItem, doc: DoclingDocument):
+    return (
+        <li>
+            <b>{node.marker}</b>
+            <span class={_text_classes(node)}>{escape(node.text)}</span>
+        </li>
+    )
+
+
+@component(CodeItem)
+def CodeComponent(children, node: CodeItem, doc: DoclingDocument):
+    return (
+        <figure>
+            <code class={_text_classes(node)}>
+                {escape(node.text or node.orig)}
+            </code>
+        </figure>
+    )
+
+
+@component(FormulaItem)
+def FormulaComponent(children, node: FormulaItem, doc: DoclingDocument):
+    return (
+        <figure>
+            <code class={_text_classes(node)}>
+                {escape(node.text or node.orig)}
+            </code>
+        </figure>
+    )
+
+
+@component(PictureItem)
+def PictureComponent(children, node: PictureItem, doc: DoclingDocument):
+    return <figure><img src={f"document/{node.self_ref[2:]}"} loading="lazy" /></figure>
+
+
+@component(PictureClassificationData)
+def PictureClassificationComponent(children, annotation: PictureClassificationData):
+    return (
+        <table>
+            <tbody>
+                {[
+                    <tr>
+                        <td>{cls.class_name.replace("_", " ")}</td>
+                        <td>{f"{cls.confidence:.2f}"}</td>
+                    </tr>
+                    for cls in annotation.predicted_classes
+                    if cls.confidence > 0.01
+                ]}
+            </tbody>
+        </table>
+    )
+
+
+@component(DescriptionAnnotation)
+def DescriptionAnnotation(children, annotation: DescriptionAnnotation):
+    return <span>{escape(annotation.text)}</span>
+
+
+@component(TableItem)
+def TableComponent(children, node: TableItem, doc: DoclingDocument):
+    covered_cells: set[(int, int)] = set()
+
+    def check_cover(cell: TableCell):
+        is_covered = (cell.start_col_offset_idx, cell.start_row_offset_idx) in covered_cells
+
+        if not is_covered:
+            for x in range(cell.start_col_offset_idx, cell.end_col_offset_idx):
+                for y in range(cell.start_row_offset_idx, cell.end_row_offset_idx):
+                    covered_cells.add((x, y))
+        
+        return is_covered
+
+    def Cell(children, cell: TableCell):
+        id = f"{node.self_ref[2:]}/{cell.start_col_offset_idx}/{cell.start_row_offset_idx}"
+
+        return (
+            <td
+                id={id}
+                class={"header" if cell.column_header or cell.row_header else None}
+                colspan={cell.col_span or 1}
+                rowspan={cell.row_span or 1}
+                onclick="clickId(event)"
+            >
+                {escape(cell.text)}
+            </td>
+        )
+
+    return (
+        <div class="table">
+            <table>
+                <tbody>
+                    {[
+                        <tr>
+                            {[
+                                <Cell cell={cell} />
+                                for cell in row
+                                if not check_cover(cell)
+                            ]}
+                        </tr>
+                        for row in node.data.grid
+                    ]}
+                </tbody>
+            </table>
+        </div>
+    )
--- a/docling_serve/ui/static/logo.svg
+++ b/docling_serve/ui/static/logo.svg
@@ -0,0 +1,116 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg width="100%" height="100%" viewBox="0 0 1024 1024" version="1.1" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" xml:space="preserve" xmlns:serif="http://www.serif.com/" style="fill-rule:evenodd;clip-rule:evenodd;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:1.5;">
+    <g id="Docling" transform="matrix(1.07666,0,0,1.07666,-35.9018,-84.1562)">
+        <g id="Outline" transform="matrix(1,0,0,1,-0.429741,55.0879)">
+            <path d="M394.709,69.09C417.34,35.077 467.97,30.178 478.031,55.609C486.35,55.043 494.726,54.701 503.158,54.589C533.157,45.238 560.496,47.419 584.65,60.732C800.941,96.66 966.069,284.814 966.069,511.232C966.069,763.284 761.435,967.918 509.383,967.918C433.692,967.918 362.277,949.464 299.385,916.808L242.3,931.993C203.092,943.242 187.715,928.369 208.575,891.871C208.935,891.24 216.518,879.37 223.997,867.677C119.604,783.975 52.698,655.355 52.698,511.232C52.698,298.778 198.086,120.013 394.709,69.09Z" style="fill:white;"/>
+        </g>
+        <g id="Color" transform="matrix(1.02317,0,0,1.02317,-11.55,-17.8333)">
+            <path d="M284.8,894.232L179.735,783.955L130.222,645.203L125.538,504.726L185.211,385.816C209.006,322.738 249.951,278.973 302.281,248.028L406.684,203.333L413.483,175.767L436.637,152.428L451.408,153.312L457.726,183.183L485.164,165.379L526.92,159.699L557.014,177.545L612.652,211.018C679.009,226.066 740.505,264.146 797.138,325.26L862.813,423.477L891.583,560.826L883.273,683.32L814.268,809.924L734.431,894.384L644.495,926.906L497.146,954.121L361.064,940.647L284.8,894.232Z" style="fill:url(#_Linear1);"/>
+            <path d="M699.932,887.255L634.427,825.291L597.884,782.352L594.906,738.956L610.14,709.396L643.207,699.954L685,710.111L730.425,736.425L765.204,778.79L775.166,849.531L719.381,894.082L699.932,887.255Z" style="fill:url(#_Linear2);"/>
+            <g transform="matrix(-0.765945,0,0,1,839.727,5.47434)">
+                <clipPath id="_clip3">
+                    <path d="M699.932,887.255L634.427,825.291L597.884,782.352L594.906,738.956L610.14,709.396L643.207,699.954L685,710.111L730.425,736.425L765.204,778.79L775.166,849.531L719.381,894.082L699.932,887.255Z"/>
+                </clipPath>
+                <g clip-path="url(#_clip3)">
+                    <g transform="matrix(-1.18516,0,0,0.907769,1039.04,88.3496)">
+                        <use xlink:href="#_Image4" x="223.969" y="674.21" width="152.098px" height="213.852px" transform="matrix(0.994105,0,0,0.999308,0,0)"/>
+                    </g>
+                </g>
+            </g>
+            <path d="M311.699,713.521C189.178,639.091 164.299,526.77 191.824,394.113L135.136,476.434L122.004,547.53C143.022,614.014 174.522,676.199 225.005,730.598C210.601,754.156 201.894,776.601 197.955,798.114L245.803,841.67C247.274,812.1 254.934,783.047 270.614,754.664L311.699,713.521Z" style="fill-opacity:0.22;"/>
+            <g transform="matrix(-1,0,0,1,1022.04,2.74442)">
+                <path d="M311.699,713.521C189.178,639.091 164.299,526.77 191.824,394.113L135.136,476.434L122.004,547.53C143.022,614.014 174.522,676.199 225.005,730.598C210.601,754.156 201.894,776.601 197.955,798.114L245.803,841.67C247.274,812.1 254.934,783.047 270.614,754.664L311.699,713.521Z" style="fill-opacity:0.22;"/>
+            </g>
+            <path d="M354.92,650.818L420.009,663.185L493.368,666.379L554.826,665.251L620.19,658.511L658.169,651.428L671.428,644.802L673.265,627.093L659.898,611.845L625.422,609.244L599.275,591.212L568.632,556.79L542.9,534.336L515.052,528.253L480.412,532.71L455.2,552.337L428.514,578.155L405.312,599.359L374.228,612.097L355.342,614.456L340.75,630.308L341.568,645.341L354.92,650.818Z" style="fill:url(#_Linear5);"/>
+            <path d="M257.168,949.32L317.434,876.747L364.928,810.6L384.1,743.934L378.759,714.719L376.844,685.849L374.836,659.954L448.734,664.2L511.462,667.602L571.339,665.091L632.796,658.836L648.232,656.882L649.937,697.808L608.105,717.702L598.45,738.594L592.286,761.642L604.743,796.309L639.595,825.803L649.872,840.757L558.219,895.152L502.124,907.569L425.781,923.496L333.29,931.298L286.269,936.907L257.168,949.32Z" style="fill:url(#_Linear6);"/>
+            <g transform="matrix(1,0,0,1.30081,-1.77636e-15,-196.488)">
+                <path d="M374.165,685.268C463.946,706.599 553.728,707.491 643.51,688.593L641.903,653.199C549.263,671.731 459.645,672.22 373.059,654.611L374.165,685.268Z" style="fill-opacity:0.18;"/>
+            </g>
+            <path d="M459.633,571.457C476.7,536.091 530.064,535.913 553.1,568.767C520.703,551.407 489.553,552.374 459.633,571.457Z" style="fill:white;"/>
+            <g transform="matrix(1,0,0,1,0.223468,-2.61949)">
+                <path d="M355.3,267.232C500.64,173.156 720.699,241.362 793.691,423.582C766.716,384.84 735.725,357.078 697.53,349.014L717.306,335.248C698.537,321.49 675.794,320.957 651.039,327.119C652.235,315.768 658.995,306.991 674.188,302.115C641.864,287.427 617.356,289.473 596.258,298.818C597.049,286.116 605.827,278.087 620.068,273.254C589.192,267.477 564.13,270.926 544.651,283.232C545.822,271.831 550.709,260.943 560.913,250.79C517.498,257.095 492.995,267.925 482.892,282.202C477.311,269.499 477.274,257.221 487.625,245.739C439.161,252.932 421.555,265.094 410.355,278.286C407.697,269.01 407.705,260.632 410.853,253.316C389.633,254.773 372.178,260.663 355.3,267.232Z" style="fill:rgb(255,213,95);"/>
+            </g>
+            <path d="M475.656,209.175C479.639,175.037 503.437,173.299 532.412,180.026C507.242,183.404 486.969,195.251 473.705,219.215L475.656,209.175Z" style="fill:rgb(255,215,101);"/>
+            <g transform="matrix(0.114323,-0.655229,0.82741,0.144365,224.632,497.317)">
+                <path d="M475.656,209.175C479.639,175.037 503.437,173.299 532.412,180.026C507.242,183.404 486.969,195.251 473.705,219.215L475.656,209.175Z" style="fill:rgb(255,215,101);"/>
+            </g>
+            <g transform="matrix(1.6739,1.15217e-16,-1.15217e-16,-0.733075,-341.46,1039.77)">
+                <path d="M447.449,560.911C468.179,536.963 546.237,539.305 565.638,560.831C533.166,555.541 477.296,553.494 447.449,560.911Z" style="fill:white;"/>
+            </g>
+            <path d="M348.201,622.341C395.549,653.534 622.351,660.854 661.936,616.729L677.568,633.834L667.044,650.308L557.802,667.518L498.074,670.562L446.718,666.416L391.404,658.406L348.154,652.501L340.161,637.119L348.201,622.341Z" style="fill:rgb(199,68,6);"/>
+        </g>
+        <g id="Black-outline" serif:id="Black outline" transform="matrix(1.02317,0,0,1.02317,-11.55,-17.8333)">
+            <path d="M373.389,657.919C376.285,676.334 377.04,695.016 375.326,714.008" style="fill:none;stroke:black;stroke-width:15.73px;"/>
+            <path d="M645.931,654.961C646.158,669.958 647.22,684.853 648.975,699.661" style="fill:none;stroke:black;stroke-width:15.73px;"/>
+            <path d="M290.084,534.662C276.554,533.535 264.892,530.024 254.279,525.175C276.732,555.341 305.316,569.76 338.631,572.029L290.084,534.662Z"/>
+            <g transform="matrix(0.94177,0,0,0.94909,28.8868,3.79501)">
+                <ellipse cx="338.022" cy="510.34" rx="88.911" ry="89.412"/>
+            </g>
+            <g transform="matrix(0.112099,0.0552506,-0.0673118,0.136571,455.367,509.409)">
+                <ellipse cx="338.022" cy="510.34" rx="88.911" ry="89.412"/>
+            </g>
+            <g transform="matrix(-0.112099,0.0552506,0.0673118,0.136571,560.529,509.492)">
+                <ellipse cx="338.022" cy="510.34" rx="88.911" ry="89.412"/>
+            </g>
+            <g transform="matrix(-1,0,0,1,1013.33,-1.15187)">
+                <path d="M290.084,534.662C276.554,533.535 264.892,530.024 254.279,525.175C276.732,555.341 305.316,569.76 338.631,572.029L290.084,534.662Z"/>
+            </g>
+            <g transform="matrix(-0.94177,0,0,0.94909,984.44,2.64314)">
+                <ellipse cx="338.022" cy="510.34" rx="88.911" ry="89.412"/>
+            </g>
+            <g transform="matrix(1,0,0,1,1.9047,-5.57346)">
+                <path d="M277.021,489.604C279.828,554.545 355.855,583.508 405.306,537.851C354.458,599.537 263.881,560.914 277.021,489.604Z" style="fill:white;"/>
+            </g>
+            <g transform="matrix(-1,0,0,1,1011.43,-5.7284)">
+                <path d="M277.021,489.604C279.828,554.545 355.855,583.508 405.306,537.851C354.458,599.537 263.881,560.914 277.021,489.604Z" style="fill:white;"/>
+            </g>
+            <g transform="matrix(0.973815,0,0,1.00246,4.71761,-0.508759)">
+                <path d="M407.22,206.891C107.655,339.384 134.447,630.03 314.615,708.305" style="fill:none;stroke:black;stroke-width:29.39px;"/>
+            </g>
+            <g transform="matrix(-0.973815,0,0,1.00246,1006.67,-1.31695)">
+                <path d="M461.559,196.756C119.768,256.762 111.059,642.544 320.305,711.486" style="fill:none;stroke:black;stroke-width:29.39px;"/>
+            </g>
+            <g id="vector-duck" serif:id="vector duck">
+                <path d="M240.912,850.71C248.043,740.231 325.609,685.992 371.268,715.193C386.487,724.926 392.506,757.72 358.575,816.753C327.005,871.68 300.465,894.596 288.329,903.447" style="fill:none;stroke:black;stroke-width:21.79px;"/>
+                <path d="M638.382,843.426C427.991,964.695 389.022,902.942 251.512,947.641L307.759,889.573" style="fill:none;stroke:black;stroke-width:15.73px;"/>
+                <path d="M770.991,853.754C779.364,764.998 730.67,727.923 666.385,704.966C629.568,691.819 580.483,723.886 595.974,772.596C606.285,805.016 650.54,839.029 707.786,886.778" style="fill:none;stroke:black;stroke-width:21.79px;"/>
+                <g transform="matrix(1,0,0,1,-1.87208,0.908099)">
+                    <path d="M603.287,772.415C614.237,757.963 627.553,750.285 642.878,748.352C628.356,760.968 617.23,775.676 620.632,799.336C635.815,785.15 650.367,779.457 664.396,780.801C651.715,790.7 639.329,803.279 641.039,818.089C641.247,819.891 647.043,823.996 647.595,825.837C659.897,816.37 672.867,811.065 689.234,809.472C676.577,822.659 668.021,834.011 674.478,848.729L664.333,847.825L625.643,812.604L603.629,786.218L603.287,772.415Z"/>
+                </g>
+                <g transform="matrix(-0.969851,0.2437,0.2437,0.969851,773.329,-138.212)">
+                    <path d="M603.287,772.415C614.237,757.963 627.553,750.285 642.878,748.352C628.356,760.968 617.23,775.676 620.632,799.336C635.815,785.15 650.367,779.457 664.396,780.801C651.715,790.7 639.329,803.279 641.039,818.089C641.247,819.891 647.043,823.996 647.595,825.837C659.897,816.37 672.867,811.065 689.234,809.472C676.577,822.659 668.021,834.011 674.478,848.729L664.333,847.825L625.643,812.604L603.629,786.218L603.287,772.415Z"/>
+                </g>
+                <path d="M511.787,670.044C461.061,671.835 411.878,662.84 361.322,653.92C329.071,648.229 335.56,616.432 361.693,615.181C391.498,613.754 411.83,601.737 437.593,569.084C459.063,541.872 482.443,528.143 506.834,529.767" style="fill:none;stroke:black;stroke-width:15.73px;"/>
+                <g transform="matrix(-1,0,0,1,1014.44,-0.213451)">
+                    <path d="M511.787,670.044C461.061,671.835 411.878,662.84 361.322,653.92C329.071,648.229 335.56,616.432 361.693,615.181C391.498,613.754 411.83,601.737 437.593,569.084C459.063,541.872 482.443,528.143 506.834,529.767" style="fill:none;stroke:black;stroke-width:15.73px;"/>
+                </g>
+            </g>
+            <g transform="matrix(2.4586,0,0,2.5497,-444.527,-690.434)">
+                <ellipse cx="312.566" cy="450.751" rx="10.63" ry="10.48" style="fill:white;"/>
+            </g>
+            <g transform="matrix(2.4586,0,0,2.5497,-127.75,-690.991)">
+                <ellipse cx="312.566" cy="450.751" rx="10.63" ry="10.48" style="fill:white;"/>
+            </g>
+            <path d="M505.738,698.061L578.879,713.989" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M422.781,709.6L568.438,743.041" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M419.941,738.409L565.688,772.989" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M408.6,787.08L510.634,810.689" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M397.571,815.956L500.93,840.219" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M386.763,844.926L454.065,861.974" style="fill:none;stroke:black;stroke-width:12.1px;"/>
+            <path d="M459.169,919.169C512.194,898.262 539.171,867.298 535.241,824.402C568.052,818.31 598.499,817.058 625.84,822.165" style="fill:none;stroke:black;stroke-width:16.95px;"/>
+            <path d="M366.219,241.106C389.605,229.261 413.371,220.601 438.247,217.5C416.795,202.419 418.72,174.582 444.22,162.47C442.086,178.175 447.633,193.354 464.772,207.738C468.721,167.57 530.015,162.087 545.674,184.112C526.45,189.314 513.082,197.344 504.566,207.717C522.403,208.119 540.706,207.86 556.2,210.609L566.935,168.471C536.388,146.208 495.718,142.166 464.65,166.705C467.703,133.264 419.536,128.364 404.624,178.47L366.219,241.106Z"/>
+            <path d="M392.617,924.576C428.953,936.938 467.84,943.636 508.258,943.636C708.944,943.636 871.876,778.49 871.876,575.076C871.876,382.463 725.788,224.162 539.898,207.895L554.137,173.696L554.485,168.187C757.218,191.602 914.895,366.003 914.895,577.383C914.895,804.698 732.549,989.249 507.949,989.249C435.381,989.249 367.223,969.983 308.199,936.232L392.617,924.576ZM279.206,917.988C171.663,843.819 101.002,718.887 101.002,577.383C101.002,383.006 234.333,219.898 413.398,176.712L424.375,216.389C264.082,254.803 144.64,400.913 144.64,575.076C144.64,703.735 209.822,817.086 308.514,883.023L279.206,917.988Z"/>
+            <path d="M714.938,895.223L647.287,836.693L616.06,855.308L549.158,889.412L459.845,919.216L390.213,928.828L429.291,950.712L535.832,960.1L586.137,952.591L662.254,931.896L714.938,895.223Z"/>
+            <path d="M423.538,929.39C509.164,917.593 580.815,890.465 640.827,850.566C635.677,886.828 622.639,918.218 594.006,939.977C530.254,930.953 474.955,928.632 423.538,929.39Z" style="fill:url(#_Linear7);"/>
+        </g>
+    </g>
+    <defs>
+        <linearGradient id="_Linear1" x1="0" y1="0" x2="1" y2="0" gradientUnits="userSpaceOnUse" gradientTransform="matrix(-52.3962,375.121,-375.121,-52.3962,471.134,384.463)"><stop offset="0" style="stop-color:rgb(255,176,44);stop-opacity:1"/><stop offset="1" style="stop-color:rgb(255,73,2);stop-opacity:1"/></linearGradient>
+        <linearGradient id="_Linear2" x1="0" y1="0" x2="1" y2="0" gradientUnits="userSpaceOnUse" gradientTransform="matrix(28.6198,-84.8913,84.8913,28.6198,647.831,831.55)"><stop offset="0" style="stop-color:rgb(255,73,2);stop-opacity:1"/><stop offset="1" style="stop-color:rgb(255,176,44);stop-opacity:1"/></linearGradient>
+        <image id="_Image4" width="153px" height="214px" xlink:href="data:image/jpeg;base64,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"/>
+        <linearGradient id="_Linear5" x1="0" y1="0" x2="1" y2="0" gradientUnits="userSpaceOnUse" gradientTransform="matrix(-39.3403,137.423,-137.423,-39.3403,545.523,573.246)"><stop offset="0" style="stop-color:rgb(255,200,41);stop-opacity:1"/><stop offset="1" style="stop-color:rgb(255,73,2);stop-opacity:1"/></linearGradient>
+        <linearGradient id="_Linear6" x1="0" y1="0" x2="1" y2="0" gradientUnits="userSpaceOnUse" gradientTransform="matrix(1.01113,-68.2054,68.2054,1.01113,482.996,741.463)"><stop offset="0" style="stop-color:white;stop-opacity:1"/><stop offset="1" style="stop-color:rgb(179,179,179);stop-opacity:1"/></linearGradient>
+        <linearGradient id="_Linear7" x1="0" y1="0" x2="1" y2="0" gradientUnits="userSpaceOnUse" gradientTransform="matrix(-7.13599,-34.117,34.117,-7.13599,578.793,922.144)"><stop offset="0" style="stop-color:rgb(164,164,164);stop-opacity:1"/><stop offset="1" style="stop-color:rgb(106,106,106);stop-opacity:1"/></linearGradient>
+    </defs>
+</svg>
--- a/docling_serve/ui/static/main.js
+++ b/docling_serve/ui/static/main.js
@@ -0,0 +1,115 @@
+// Propagate URL hash to CSS target class for elements with the same id or data-id.
+window.addEventListener("hashchange", function (event) {
+  [
+    ["remove", "oldURL"],
+    ["add", "newURL"],
+  ].forEach(([op, tense]) => {
+    const hash = new URL(event[tense]).hash.slice(1);
+    document
+      .querySelectorAll(`[data-id="${hash}"], [id="${hash}"]`)
+      .forEach((el) => el.classList[op]("target"));
+  });
+});
+
+// Navigate document items with cursor keys.
+document.addEventListener("keydown", function (event) {
+  const target = document.querySelector("*:target");
+  const tbounds = target?.getBoundingClientRect();
+  const filters = {
+    ArrowUp: (_x, y) => y < tbounds.top,
+    ArrowDown: (_x, y) => y > tbounds.bottom,
+    ArrowLeft: (x, _y) => x < tbounds.left,
+    ArrowRight: (x, _y) => x > tbounds.right,
+  };
+
+  if (target && filters[event.key]) {
+    const elements = [...document.querySelectorAll(".item[id], .item *[id]")];
+
+    let minEl, minDist;
+    for (const el of elements) {
+      const elBounds = el.getBoundingClientRect();
+
+      if (
+        filters[event.key](
+          (elBounds.left + elBounds.right) / 2,
+          (elBounds.top + elBounds.bottom) / 2
+        )
+      ) {
+        const elDist =
+          Math.abs(tbounds.x - elBounds.x) + Math.abs(tbounds.y - elBounds.y);
+
+        if (el != target && elDist < (minDist ?? Number.MAX_VALUE)) {
+          minEl = el;
+          minDist = elDist;
+        }
+      }
+    }
+
+    if (minEl) {
+      event.preventDefault();
+      location.href = `#${minEl.id}`;
+    }
+  }
+});
+
+// Navigate to item with id when it is clicked.
+function clickId(e) {
+  e.stopPropagation();
+  const id = e.currentTarget.getAttribute("data-id") ?? e.currentTarget.id;
+  location.href = `#${id}`;
+}
+
+window.onload = () => {
+  // (Re-)set the value of input[data-dep-on] to conform to a value of another input[name="data-dep-on"].
+  document.querySelectorAll("input[dep-on]").forEach((element) => {
+    const onName = element.getAttribute("dep-on");
+    const onElement = document.getElementsByName(onName)[0];
+    const depMap = JSON.parse(element.getAttribute("dep-values") ?? "{}");
+
+    if (onElement && depMap) {
+      // On load.
+      element.value = depMap[onElement.value] ?? "";
+
+      // On change.
+      onElement.addEventListener(
+        "change",
+        (event) => (element.value = depMap[event.currentTarget.value] ?? "")
+      );
+    }
+  });
+
+  // Toggle display of input[data-display-when] when it requires a different input[type=checkbox] to be checked.
+  document.querySelectorAll("*[display-when]").forEach((element) => {
+    const whenElements = element
+      .getAttribute("display-when")
+      .split(",")
+      .flatMap((whenName) => [...document.getElementsByName(whenName.trim())]);
+
+    function update() {
+      const allChecked = whenElements.every((el) => el.checked);
+      element.classList[allChecked ? "remove" : "add"]("hidden");
+    }
+
+    // On load.
+    update();
+
+    // On change.
+    whenElements.forEach((whenElement) =>
+      whenElement.addEventListener("change", update)
+    );
+  });
+
+  // Persist input value in local storage.
+  document
+    .querySelectorAll("input[type=checkbox][persist]")
+    .forEach((element) => {
+      const prefix = element.getAttribute("persist");
+      const name = element.getAttribute("name");
+      const key = `docling-serve-${prefix}-${name}`;
+
+      element.checked = localStorage.getItem(key) === "true";
+      element.addEventListener("change", (event) =>
+        localStorage.setItem(key, event.target.checked)
+      );
+    });
+};
--- a/docling_serve/ui/static/pico.css
+++ b/docling_serve/ui/static/pico.css
--- a/docling_serve/ui/static/style.css
+++ b/docling_serve/ui/static/style.css
@@ -0,0 +1,429 @@
+@import "pico.css";
+
+@view-transition {
+  navigation: auto;
+}
+
+:root {
+  --pico-font-size: 16px;
+
+  --highlight-factor: 0.8;
+  --target: hsl(240, 100%, 34%);
+  --mark: hsl(29, 100%, 35%);
+}
+
+@media (prefers-color-scheme: dark) {
+  :root {
+    --highlight-factor: 1.5;
+    --target: hsl(240, 100%, 70%);
+    --mark: hsl(29, 100%, 70%);
+  }
+}
+
+/* Utilities. */
+.hidden {
+  display: none;
+}
+
+.sticky-footer {
+  position: sticky;
+  bottom: 0;
+  padding-top: var(--pico-spacing);
+  background: var(--pico-background-color);
+}
+
+html {
+  scroll-behavior: smooth;
+}
+
+header {
+  position: relative;
+  display: flex;
+  gap: 5rem;
+  margin-bottom: 2rem;
+
+  > .title {
+    white-space: nowrap;
+    font-size: 2rem;
+    font-weight: 300;
+    line-height: 1.75;
+
+    img {
+      display: inline-block;
+      max-height: 0.8em;
+      margin: -0.2rem -0.2em 0.25rem -0.2em;
+    }
+  }
+
+  &.loading img {
+    animation: shake 0.5s ease-in-out alternate infinite;
+    scale: 1.5;
+    translate: 0 1.5rem;
+  }
+
+  > .version {
+    position: absolute;
+    left: 6.25rem;
+    bottom: -0.5rem;
+    padding: 0 0.25rem;
+    font-size: 0.65rem;
+    line-height: 1rem;
+    border: solid 1px var(--pico-color);
+    border-radius: 0.3rem;
+  }
+
+  @media (prefers-color-scheme: dark) {
+    --glow: hsl(29, 100%, 70%);
+
+    > .title {
+      text-shadow: 0 0 0.25rem white, 0 0 0.5rem var(--glow),
+        0 0 0.75rem var(--glow), 0 0 1rem var(--glow);
+      color: white;
+
+      img {
+        filter: drop-shadow(0 0 0.05rem white)
+          drop-shadow(0 0 0.1rem var(--glow))
+          drop-shadow(0 0 0.15rem var(--glow))
+          drop-shadow(0 0 0.2rem var(--glow));
+      }
+    }
+
+    > .version {
+      color: white;
+      border-color: white;
+      text-shadow: 0 0 0.05rem white, 0 0 0.1rem var(--glow),
+        0 0 0.15rem var(--glow), 0 0 0.2rem var(--glow);
+      box-shadow: 0 0 0.05rem white, 0 0 0.1rem var(--glow),
+        0 0 0.15rem var(--glow), 0 0 0.2rem var(--glow);
+    }
+  }
+}
+
+@keyframes shake {
+  50% {
+    transform: rotate(-20deg);
+  }
+  100% {
+    transform: rotate(20deg);
+  }
+}
+
+label > small {
+  margin-left: var(--pico-spacing);
+  opacity: 0.75;
+}
+
+/* Conversion results. */
+.progress,
+.fail {
+  margin-top: 3rem;
+}
+
+.fail {
+  color: var(--pico-del-color);
+}
+
+main.preview {
+  display: grid;
+  grid:
+    auto / 1fr 0.5rem minmax(20ch, 70ch) 0.5rem minmax(min-content, auto)
+    minmax(0.5rem, 1fr);
+  grid-auto-flow: dense;
+  align-content: start;
+}
+
+/* Header and task status. */
+main.preview {
+  > header {
+    grid-column: 3;
+    padding: 0 0.5rem;
+  }
+
+  > .status {
+    grid-row: 2;
+    grid-column: 3;
+    display: inline-block;
+    margin: 0 0.5rem 3rem 0.5rem;
+
+    span {
+      display: inline-block;
+      min-width: calc(5 * var(--pico-spacing));
+      padding-right: calc(0.5 * var(--pico-spacing));
+    }
+  }
+
+  > .formats {
+    grid-row: 2;
+    grid-column: 5;
+    margin-bottom: 3rem;
+    display: flex;
+    align-items: flex-end;
+    gap: 1rem;
+
+    > .configDarkImg {
+      display: none;
+      grid-row: 2;
+      grid-column: 6;
+      margin-left: auto;
+    }
+
+    @media (prefers-color-scheme: dark) {
+      > .configDarkImg {
+        display: block;
+      }
+    }
+  }
+}
+
+/* Invert images in dark mode (option). */
+@media (prefers-color-scheme: dark) {
+  main.preview:has(.configDarkImg > input:checked) {
+    --img-hover-border: white;
+
+    svg.page-image {
+      --mark: hsl(29, 100%, 70%)
+    }
+
+    image,
+    img {
+      filter: invert(1) hue-rotate(180deg) saturate(1.25);
+    }
+  }
+}
+
+/* Document contents. */
+main.preview {
+  --img-hover-border: black;
+
+  *[id] {
+    scroll-margin-top: 20vh;
+  }
+
+  > .page {
+    position: relative;
+    display: grid;
+    grid-template-columns: subgrid;
+    grid-auto-flow: dense;
+    grid-column: 1 / span 6;
+
+    > .item {
+      grid-column: 3;
+      width: 100%;
+      min-height: 3rem;
+      max-height: fit-content;
+      margin: 0;
+      padding: 0.5rem;
+      text-align: justify;
+      background-color: var(--pico-background-color);
+
+      cursor: pointer;
+
+      &:hover {
+        filter: brightness(var(--highlight-factor));
+      }
+
+      &.target {
+        outline: 2px solid var(--target);
+        z-index: 10;
+      }
+    }
+
+    > .item.void {
+      visibility: hidden;
+    }
+
+    > .item.annotated {
+      display: flex;
+      flex-direction: column;
+      align-items: stretch;
+      gap: 1rem;
+    }
+
+    /* Formatting. */
+    .bold {
+      font-weight: bold;
+    }
+    .italic {
+      font-style: italic;
+    }
+    .underline {
+      text-decoration: underline;
+    }
+    .strikethrough {
+      text-decoration: line-through;
+    }
+    .underline.strikethrough {
+      text-decoration: underline line-through;
+    }
+    .sub {
+      font-size: smaller;
+      vertical-align: sub;
+    }
+    .super {
+      font-size: smaller;
+      vertical-align: super;
+    }
+
+    /* Items out of content layer. */
+    > .item:not(.body),
+    > .item-markers:not(.body) {
+      opacity: 0.5;
+    }
+
+    > li.item {
+      list-style-type: none;
+    }
+
+    > .item.caption {
+      padding: 0.5rem 1.5rem;
+      font-size: 0.9rem;
+    }
+
+    > .item.table {
+      min-width: 0;
+      overflow-x: auto;
+
+      table {
+        font-size: 0.75rem;
+        border-collapse: collapse;
+
+        td {
+          vertical-align: top;
+        }
+
+        td.header {
+          font-weight: bold;
+          background-color: var(--pico-code-background-color);
+        }
+
+        td.target {
+          outline: solid 2px var(--target);
+        }
+      }
+    }
+
+    .annotation {
+      margin: 0;
+
+      &::before {
+        content: attr(data-kind);
+        opacity: 0.7;
+      }
+
+      &,
+      * {
+        font-size: 0.9rem;
+        color: var(--mark);
+      }
+    }
+
+    .annotation[data-kind="description"],
+    code.annotation {
+      white-space: pre-line;
+    }
+
+    .annotation[data-kind="classification"] {
+      width: fit-content;
+    }
+
+    > .item-markers {
+      position: relative;
+      grid-column: 2;
+      padding-top: 0.125rem;
+      padding-right: 0.5rem;
+
+      display: flex;
+      flex-direction: column;
+      align-items: flex-end;
+
+      font-family: monospace;
+      font-size: 0.675rem;
+      line-height: 1.25;
+      letter-spacing: 0;
+      color: var(--mark);
+      white-space: nowrap;
+      border-top: solid 1px var(--mark);
+
+      > * {
+        margin-right: 0.5rem;
+      }
+
+      > a {
+        padding: 0.125rem 0.25rem;
+        margin-top: 0.5rem;
+        border-radius: 0.125rem;
+        color: var(--pico-contrast-inverse);
+        background-color: var(--target);
+        text-decoration: none;
+      }
+
+      > a:hover {
+        filter: brightness(--highlight-factor);
+      }
+
+      &:not(.target) > a {
+        display: none;
+      }
+
+      &.group,
+      &.grouped {
+        border-left: 1px dashed var(--mark);
+      }
+
+      &.group {
+        margin-top: 0.5rem;
+        border-top: none;
+      }
+    }
+
+    > .page-marker {
+      grid-column: 4;
+
+      &.border {
+        transform: translateY(-1px);
+        border-top: solid 1px var(--mark);
+      }
+    }
+
+    > svg.page-image {
+      --mark: hsl(29, 100%, 35%);
+
+      grid-column: 5;
+      position: sticky;
+      top: 0.5rem;
+      width: 100%;
+      max-height: calc(100vh - 1rem);
+      outline: 1px solid var(--mark);
+
+      rect {
+        stroke: var(--mark);
+        stroke-width: 1px;
+        fill: var(--target);
+        fill-opacity: 0.0001; /* To activate hover. */
+        cursor: pointer;
+
+        &:hover {
+          filter: brightness(0.8);
+          fill-opacity: 0.1;
+          stroke: var(--img-hover-border);
+          stroke-width: 3px;
+        }
+      }
+
+      rect.target {
+        stroke: var(--target);
+        stroke-width: 3px;
+        stroke-dasharray: none;
+      }
+
+      text {
+        font-size: 0.675rem;
+        color: var(--mark);
+
+        &.top-no {
+          alignment-baseline: hanging;
+        }
+      }
+    }
+  }
+}
--- a/docling_serve/ui/svg.py
+++ b/docling_serve/ui/svg.py
@@ -0,0 +1,20 @@
+from pyjsx import JSX  # type: ignore
+
+
+def _tag(name: str):
+    def factory(children, **args) -> JSX:
+        props = " ".join([f'{k}="{v}"' for k, v in args.items()])
+
+        if children:
+            child_renders = "".join([str(c) for c in children])
+            return f"<{name} {props}>{child_renders}</{name}>"
+        else:
+            return f"<{name} {props} />"
+
+    return factory
+
+
+image = _tag("image")
+path = _tag("path")
+rect = _tag("rect")
+text = _tag("text")
--- a/docling_serve/websocket_notifier.py
+++ b/docling_serve/websocket_notifier.py
@@ -30,26 +30,47 @@ class WebsocketNotifier(BaseNotifier):
        if task_id not in self.task_subscribers:
            raise RuntimeError(f"Task {task_id} does not have a subscribers list.")

-        task = await self.orchestrator.get_raw_task(task_id=task_id)
-        task_queue_position = await self.orchestrator.get_queue_position(task_id)
-        msg = TaskStatusResponse(
-            task_id=task.task_id,
-            task_type=task.task_type,
-            task_status=task.task_status,
-            task_position=task_queue_position,
-            task_meta=task.processing_meta,
-        )
-        for websocket in self.task_subscribers[task_id]:
-            await websocket.send_text(
-                WebsocketMessage(message=MessageKind.UPDATE, task=msg).model_dump_json()
+        try:
+            # Get task status from Redis or RQ directly instead of in-memory registry
+            task = await self.orchestrator.task_status(task_id=task_id)
+            task_queue_position = await self.orchestrator.get_queue_position(task_id)
+            msg = TaskStatusResponse(
+                task_id=task.task_id,
+                task_type=task.task_type,
+                task_status=task.task_status,
+                task_position=task_queue_position,
+                task_meta=task.processing_meta,
            )
-            if task.is_completed():
-                await websocket.close()
+            for websocket in self.task_subscribers[task_id]:
+                await websocket.send_text(
+                    WebsocketMessage(
+                        message=MessageKind.UPDATE, task=msg
+                    ).model_dump_json()
+                )
+                if task.is_completed():
+                    await websocket.close()
+        except Exception as e:
+            # Log the error but don't crash the notifier
+            import logging
+
+            _log = logging.getLogger(__name__)
+            _log.error(f"Error notifying subscribers for task {task_id}: {e}")

    async def notify_queue_positions(self):
+        """Notify all subscribers of pending tasks about queue position updates."""
        for task_id in self.task_subscribers.keys():
-            # notify only pending tasks
-            if self.orchestrator.tasks[task_id].task_status != TaskStatus.PENDING:
-                continue
+            try:
+                # Check task status directly from Redis or RQ
+                task = await self.orchestrator.task_status(task_id)

-            await self.notify_task_subscribers(task_id)
+                # Notify only pending tasks
+                if task.task_status == TaskStatus.PENDING:
+                    await self.notify_task_subscribers(task_id)
+            except Exception as e:
+                # Log the error but don't crash the notifier
+                import logging
+
+                _log = logging.getLogger(__name__)
+                _log.error(
+                    f"Error checking task {task_id} status for queue position notification: {e}"
+                )
--- a/docs/configuration.md
+++ b/docs/configuration.md
@@ -39,23 +39,42 @@ THe following table describes the options to configure the Docling Serve app.
 |  | `DOCLING_SERVE_STATIC_PATH` | unset | If set to a valid directory, the static assets for the docs and UI will be loaded from this path |
 |  | `DOCLING_SERVE_SCRATCH_PATH` |  | If set, this directory will be used as scratch workspace, e.g. storing the results before they get requested. If unset, a temporary created is created for this purpose. |
 | `--enable-ui` | `DOCLING_SERVE_ENABLE_UI` | `false` | Enable the demonstrator UI. |
+|  | `DOCLING_SERVE_SHOW_VERSION_INFO` | `true` | If enabled, the `/version` endpoint will provide the Docling package versions, otherwise it will return a forbidden 403 error. |
 |  | `DOCLING_SERVE_ENABLE_REMOTE_SERVICES` | `false` | Allow pipeline components making remote connections. For example, this is needed when using a vision-language model via APIs. |
 |  | `DOCLING_SERVE_ALLOW_EXTERNAL_PLUGINS` | `false` | Allow the selection of third-party plugins. |
 |  | `DOCLING_SERVE_SINGLE_USE_RESULTS` | `true` | If true, results can be accessed only once. If false, the results accumulate in the scratch directory. |
 |  | `DOCLING_SERVE_RESULT_REMOVAL_DELAY` | `300` | When `DOCLING_SERVE_SINGLE_USE_RESULTS` is active, this is the delay before results are removed from the task registry. |
 |  | `DOCLING_SERVE_MAX_DOCUMENT_TIMEOUT` | `604800` (7 days) | The maximum time for processing a document. |
-|  | `DOCLING_NUM_THREADS` | `4` | Number of concurrent threads for processing a document. |
 |  | `DOCLING_SERVE_MAX_NUM_PAGES` |  | The maximum number of pages for a document to be processed. |
 |  | `DOCLING_SERVE_MAX_FILE_SIZE` |  | The maximum file size for a document to be processed. |
+|  | `DOCLING_SERVE_SYNC_POLL_INTERVAL` | `2` | Number of seconds to sleep between polling the task status in the sync endpoints. |
 |  | `DOCLING_SERVE_MAX_SYNC_WAIT` | `120` | Max number of seconds a synchronous endpoint is waiting for the task completion. |
 |  | `DOCLING_SERVE_LOAD_MODELS_AT_BOOT` | `True` | If enabled, the models for the default options will be loaded at boot. |
 |  | `DOCLING_SERVE_OPTIONS_CACHE_SIZE` | `2` | How many DocumentConveter objects (including their loaded models) to keep in the cache. |
+|  | `DOCLING_SERVE_QUEUE_MAX_SIZE` | | Size of the pages queue. Potentially so many pages opened at the same time. |
+|  | `DOCLING_SERVE_OCR_BATCH_SIZE` | | Batch size for the OCR stage. |
+|  | `DOCLING_SERVE_LAYOUT_BATCH_SIZE` | | Batch size for the layout detection stage. |
+|  | `DOCLING_SERVE_TABLE_BATCH_SIZE` | | Batch size for the table structure stage. |
+|  | `DOCLING_SERVE_BATCH_POLLING_INTERVAL_SECONDS` | | Wait time for gathering pages before starting a stage processing. |
 |  | `DOCLING_SERVE_CORS_ORIGINS` | `["*"]` | A list of origins that should be permitted to make cross-origin requests. |
 |  | `DOCLING_SERVE_CORS_METHODS` | `["*"]` | A list of HTTP methods that should be allowed for cross-origin requests. |
 |  | `DOCLING_SERVE_CORS_HEADERS` | `["*"]` | A list of HTTP request headers that should be supported for cross-origin requests. |
 |  | `DOCLING_SERVE_API_KEY` | | If specified, all the API requests must contain the header `X-Api-Key` with this value. |
 |  | `DOCLING_SERVE_ENG_KIND` | `local` | The compute engine to use for the async tasks. Possible values are `local`, `rq` and `kfp`. See below for more configurations of the engines. |

+### Docling configuration
+
+Some Docling settings, mostly about performance, are exposed as environment variable which can be used also when running Docling Serve.
+
+| ENV | Default | Description |
+| ----|---------|-------------|
+| `DOCLING_NUM_THREADS` | `4` | Number of concurrent threads used for the `torch` CPU execution. |
+| `DOCLING_DEVICE` | | Device used for the model execution. Valid values are `cpu`, `cuda`, `mps`. When unset, the best device is chosen. For CUDA-enabled environments, you can choose which GPU using the syntax `cuda:0`, `cuda:1`, ... |
+| `DOCLING_PERF_PAGE_BATCH_SIZE` | `4` | Number of pages processed in the same batch. |
+| `DOCLING_PERF_ELEMENTS_BATCH_SIZE` | `8` | Number of document items/elements processed in the same batch during enrichment. |
+| `DOCLING_DEBUG_PROFILE_PIPELINE_TIMINGS` | `false` | When enabled, Docling will provide detailed timings information. |
+
+
 ### Compute engine

 Docling Serve can be deployed with several possible of compute engine.
--- a/docs/usage.md
+++ b/docs/usage.md
@@ -4,31 +4,89 @@ The API provides two endpoints: one for urls, one for files. This is necessary t

 ## Common parameters

-On top of the source of file (see below), both endpoints support the same parameters, which are almost the same as the Docling CLI.
+On top of the source of file (see below), both endpoints support the same parameters.

- `from_formats` (List[str]): Input format(s) to convert from. Allowed values: `docx`, `pptx`, `html`, `image`, `pdf`, `asciidoc`, `md`. Defaults to all formats.
- `to_formats` (List[str]): Output format(s) to convert to. Allowed values: `md`, `json`, `html`, `text`, `doctags`. Defaults to `md`.
- `pipeline` (str). The choice of which pipeline to use. Allowed values are `standard` and `vlm`. Defaults to `standard`.
- `page_range` (tuple). If specified, only convert a range of pages. The page number starts at 1.
- `do_ocr` (bool): If enabled, the bitmap content will be processed using OCR. Defaults to `True`.
- `image_export_mode`: Image export mode for the document (only in case of JSON, Markdown or HTML). Allowed values: embedded, placeholder, referenced. Optional, defaults to `embedded`.
- `force_ocr` (bool): If enabled, replace any existing text with OCR-generated text over the full content. Defaults to `False`.
- `ocr_engine` (str): OCR engine to use. Allowed values: `easyocr`, `tesserocr`, `tesseract`, `rapidocr`, `ocrmac`. Defaults to `easyocr`. To use the `tesserocr` engine, `tesserocr` must be installed where docling-serve is running: `pip install tesserocr`
- `ocr_lang` (List[str]): List of languages used by the OCR engine. Note that each OCR engine has different values for the language names. Defaults to empty.
- `pdf_backend` (str): PDF backend to use. Allowed values: `pypdfium2`, `dlparse_v1`, `dlparse_v2`, `dlparse_v4`. Defaults to `dlparse_v4`.
- `table_mode` (str): Table mode to use. Allowed values: `fast`, `accurate`. Defaults to `fast`.
- `abort_on_error` (bool): If enabled, abort on error. Defaults to false.
- `md_page_break_placeholder` (str): Add this placeholder between pages in the markdown output.
- `do_table_structure` (bool): If enabled, the table structure will be extracted. Defaults to true.
- `do_code_enrichment` (bool): If enabled, perform OCR code enrichment. Defaults to false.
- `do_formula_enrichment` (bool): If enabled, perform formula OCR, return LaTeX code. Defaults to false.
- `do_picture_classification` (bool): If enabled, classify pictures in documents. Defaults to false.
- `do_picture_description` (bool): If enabled, describe pictures in documents. Defaults to false.
- `picture_description_area_threshold` (float): Minimum percentage of the area for a picture to be processed with the models. Defaults to 0.05.
- `picture_description_local` (dict): Options for running a local vision-language model in the picture description. The parameters refer to a model hosted on Hugging Face. This parameter is mutually exclusive with `picture_description_api`.
- `picture_description_api` (dict): API details for using a vision-language model in the picture description. This parameter is mutually exclusive with `picture_description_local`.
- `include_images` (bool): If enabled, images will be extracted from the document. Defaults to false.
- `images_scale` (float): Scale factor for images. Defaults to 2.0.
+<!-- begin: parameters-docs -->
+<h4>ConvertDocumentsRequestOptions</h4>
+
+| Field Name | Type | Description |
+|------------|------|-------------|
+| `from_formats` | List[InputFormat] | Input format(s) to convert from. String or list of strings. Allowed values: `docx`, `pptx`, `html`, `image`, `pdf`, `asciidoc`, `md`, `csv`, `xlsx`, `xml_uspto`, `xml_jats`, `mets_gbs`, `json_docling`, `audio`, `vtt`. Optional, defaults to all formats. |
+| `to_formats` | List[OutputFormat] | Output format(s) to convert to. String or list of strings. Allowed values: `md`, `json`, `html`, `html_split_page`, `text`, `doctags`. Optional, defaults to Markdown. |
+| `image_export_mode` | ImageRefMode | Image export mode for the document (in case of JSON, Markdown or HTML). Allowed values: `placeholder`, `embedded`, `referenced`. Optional, defaults to Embedded. |
+| `do_ocr` | bool | If enabled, the bitmap content will be processed using OCR. Boolean. Optional, defaults to true |
+| `force_ocr` | bool | If enabled, replace existing text with OCR-generated text over content. Boolean. Optional, defaults to false. |
+| `ocr_engine` | `ocr_engines_enum` | The OCR engine to use. String. Allowed values: `auto`, `easyocr`, `ocrmac`, `rapidocr`, `tesserocr`, `tesseract`. Optional, defaults to `easyocr`. |
+| `ocr_lang` | List[str] or NoneType | List of languages used by the OCR engine. Note that each OCR engine has different values for the language names. String or list of strings. Optional, defaults to empty. |
+| `pdf_backend` | PdfBackend | The PDF backend to use. String. Allowed values: `pypdfium2`, `dlparse_v1`, `dlparse_v2`, `dlparse_v4`. Optional, defaults to `dlparse_v4`. |
+| `table_mode` | TableFormerMode | Mode to use for table structure, String. Allowed values: `fast`, `accurate`. Optional, defaults to accurate. |
+| `table_cell_matching` | bool | If true, matches table cells predictions back to PDF cells. Can break table output if PDF cells are merged across table columns. If false, let table structure model define the text cells, ignore PDF cells. |
+| `pipeline` | ProcessingPipeline | Choose the pipeline to process PDF or image files. |
+| `page_range` | Tuple | Only convert a range of pages. The page number starts at 1. |
+| `document_timeout` | float | The timeout for processing each document, in seconds. |
+| `abort_on_error` | bool | Abort on error if enabled. Boolean. Optional, defaults to false. |
+| `do_table_structure` | bool | If enabled, the table structure will be extracted. Boolean. Optional, defaults to true. |
+| `include_images` | bool | If enabled, images will be extracted from the document. Boolean. Optional, defaults to true. |
+| `images_scale` | float | Scale factor for images. Float. Optional, defaults to 2.0. |
+| `md_page_break_placeholder` | str | Add this placeholder between pages in the markdown output. |
+| `do_code_enrichment` | bool | If enabled, perform OCR code enrichment. Boolean. Optional, defaults to false. |
+| `do_formula_enrichment` | bool | If enabled, perform formula OCR, return LaTeX code. Boolean. Optional, defaults to false. |
+| `do_picture_classification` | bool | If enabled, classify pictures in documents. Boolean. Optional, defaults to false. |
+| `do_picture_description` | bool | If enabled, describe pictures in documents. Boolean. Optional, defaults to false. |
+| `picture_description_area_threshold` | float | Minimum percentage of the area for a picture to be processed with the models. |
+| `picture_description_local` | PictureDescriptionLocal or NoneType | Options for running a local vision-language model in the picture description. The parameters refer to a model hosted on Hugging Face. This parameter is mutually exclusive with `picture_description_api`. |
+| `picture_description_api` | PictureDescriptionApi or NoneType | API details for using a vision-language model in the picture description. This parameter is mutually exclusive with `picture_description_local`. |
+| `vlm_pipeline_model` | VlmModelType or NoneType | Preset of local and API models for the `vlm` pipeline. This parameter is mutually exclusive with `vlm_pipeline_model_local` and `vlm_pipeline_model_api`. Use the other options for more parameters. |
+| `vlm_pipeline_model_local` | VlmModelLocal or NoneType | Options for running a local vision-language model for the `vlm` pipeline. The parameters refer to a model hosted on Hugging Face. This parameter is mutually exclusive with `vlm_pipeline_model_api` and `vlm_pipeline_model`. |
+| `vlm_pipeline_model_api` | VlmModelApi or NoneType | API details for using a vision-language model for the `vlm` pipeline. This parameter is mutually exclusive with `vlm_pipeline_model_local` and `vlm_pipeline_model`. |
+
+<h4>VlmModelApi</h4>
+
+| Field Name | Type | Description |
+|------------|------|-------------|
+| `url` | AnyUrl | Endpoint which accepts openai-api compatible requests. |
+| `headers` | Dict[str, str] | Headers used for calling the API endpoint. For example, it could include authentication headers. |
+| `params` | Dict[str, Any] | Model parameters. |
+| `timeout` | float | Timeout for the API request. |
+| `concurrency` | int | Maximum number of concurrent requests to the API. |
+| `prompt` | str | Prompt used when calling the vision-language model. |
+| `scale` | float | Scale factor of the images used. |
+| `response_format` | ResponseFormat | Type of response generated by the model. |
+| `temperature` | float | Temperature parameter controlling the reproducibility of the result. |
+
+<h4>VlmModelLocal</h4>
+
+| Field Name | Type | Description |
+|------------|------|-------------|
+| `repo_id` | str | Repository id from the Hugging Face Hub. |
+| `prompt` | str | Prompt used when calling the vision-language model. |
+| `scale` | float | Scale factor of the images used. |
+| `response_format` | ResponseFormat | Type of response generated by the model. |
+| `inference_framework` | InferenceFramework | Inference framework to use. |
+| `transformers_model_type` | TransformersModelType | Type of transformers auto-model to use. |
+| `extra_generation_config` | Dict[str, Any] | Config from https://huggingface.co/docs/transformers/en/main_classes/text_generation#transformers.GenerationConfig |
+| `temperature` | float | Temperature parameter controlling the reproducibility of the result. |
+
+<h4>PictureDescriptionApi</h4>
+
+| Field Name | Type | Description |
+|------------|------|-------------|
+| `url` | AnyUrl | Endpoint which accepts openai-api compatible requests. |
+| `headers` | Dict[str, str] | Headers used for calling the API endpoint. For example, it could include authentication headers. |
+| `params` | Dict[str, Any] | Model parameters. |
+| `timeout` | float | Timeout for the API request. |
+| `concurrency` | int | Maximum number of concurrent requests to the API. |
+| `prompt` | str | Prompt used when calling the vision-language model. |
+
+<h4>PictureDescriptionLocal</h4>
+
+| Field Name | Type | Description |
+|------------|------|-------------|
+| `repo_id` | str | Repository id from the Hugging Face Hub. |
+| `prompt` | str | Prompt used when calling the vision-language model. |
+| `generation_config` | Dict[str, Any] | Config from https://huggingface.co/docs/transformers/en/main_classes/text_generation#transformers.GenerationConfig |
+
+<!-- end: parameters-docs -->

 ### Authentication

@@ -433,7 +491,7 @@ with connect(uri) as websocket:
            payload = json.loads(message)
            if payload["message"] == "error":
                break
-            if payload["message"] == "error" and payload["task"]["task_status"] in ("success", "failure"):
+            if payload["message"] == "update" and payload["task"]["task_status"] in ("success", "failure"):
                break
        except:
          break
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "docling-serve"
-version = "1.5.1"  # DO NOT EDIT, updated automatically
+version = "1.8.0"  # DO NOT EDIT, updated automatically
 description = "Running Docling as a service"
 license = {text = "MIT"}
 authors = [
@@ -35,8 +35,8 @@ requires-python = ">=3.10"
 dependencies = [
    "docling~=2.38",
    "docling-core>=2.45.0",
-    "docling-jobkit[kfp,rq,vlm]>=1.5.0,<2.0.0",
-    "fastapi[standard]~=0.115",
+    "docling-jobkit[kfp,rq,vlm]>=1.8.0,<2.0.0",
+    "fastapi[standard]<0.119.0",  # ~=0.115
    "httpx~=0.28",
    "pydantic~=2.10",
    "pydantic-settings~=2.4",
@@ -50,15 +50,17 @@ dependencies = [

 [project.optional-dependencies]
 ui = [
-    "gradio~=5.23.2",
+    "python-jsx>=0.2.0",
 ]
 tesserocr = [
    "tesserocr~=2.7"
 ]
+easyocr = [
+    "easyocr>=1.7",
+]
 rapidocr = [
    "rapidocr (>=3.3,<4.0.0) ; python_version < '3.14'",
    "onnxruntime (>=1.7.0,<2.0.0)",
-      "modelscope>=1.29.0",
 ]
 flash-attn = [
  "flash-attn~=2.8.2; sys_platform == 'linux' and platform_machine == 'x86_64'"
@@ -123,7 +125,8 @@ conflicts = [
 ]
 environments = ["sys_platform != 'darwin' or platform_machine != 'x86_64'"]
 override-dependencies = [
-  "urllib3~=2.0"
+  "urllib3~=2.0",
+  "xgrammar>=0.1.24"
 ]

 [tool.uv.sources]
--- a/scripts/init.py
+++ b/scripts/init.py
--- a/scripts/update_doc_usage.py
+++ b/scripts/update_doc_usage.py
@@ -0,0 +1,199 @@
+import re
+from typing import Annotated, Any, Union, get_args, get_origin
+
+from pydantic import BaseModel
+
+from docling_serve.datamodel.convert import ConvertDocumentsRequestOptions
+
+DOCS_FILE = "docs/usage.md"
+
+VARIABLE_WORDS: list[str] = [
+    "picture_description_local",
+    "vlm_pipeline_model",
+    "vlm",
+    "vlm_pipeline_model_api",
+    "ocr_engines_enum",
+    "easyocr",
+    "dlparse_v4",
+    "fast",
+    "picture_description_api",
+    "vlm_pipeline_model_local",
+]
+
+
+def format_variable_names(text: str) -> str:
+    """Format specific words in description to be code-formatted."""
+    sorted_words = sorted(VARIABLE_WORDS, key=len, reverse=True)
+
+    escaped_words = [re.escape(word) for word in sorted_words]
+
+    for word in escaped_words:
+        pattern = rf"(?<!`)\b{word}\b(?!`)"
+        text = re.sub(pattern, f"`{word}`", text)
+
+    return text
+
+
+def format_allowed_values_description(description: str) -> str:
+    """Format description to code-format allowed values."""
+    # Regex pattern to find text after "Allowed values:"
+    match = re.search(r"Allowed values:(.+?)(?:\.|$)", description, re.DOTALL)
+
+    if match:
+        # Extract the allowed values
+        values_str = match.group(1).strip()
+
+        # Split values, handling both comma and 'and' separators
+        values = re.split(r"\s*(?:,\s*|\s+and\s+)", values_str)
+
+        # Remove any remaining punctuation and whitespace
+        values = [value.strip("., ") for value in values]
+
+        # Create code-formatted values
+        formatted_values = ", ".join(f"`{value}`" for value in values)
+
+        # Replace the original allowed values with formatted version
+        formatted_description = re.sub(
+            r"(Allowed values:)(.+?)(?:\.|$)",
+            f"\\1 {formatted_values}.",
+            description,
+            flags=re.DOTALL,
+        )
+
+        return formatted_description
+
+    return description
+
+
+def _format_type(type_hint: Any) -> str:
+    """Format type ccrrectly, like Annotation or Union."""
+    if get_origin(type_hint) is Annotated:
+        base_type = get_args(type_hint)[0]
+        return _format_type(base_type)
+
+    if hasattr(type_hint, "__origin__"):
+        origin = type_hint.__origin__
+        args = get_args(type_hint)
+
+        if origin is list:
+            return f"List[{_format_type(args[0])}]"
+        elif origin is dict:
+            return f"Dict[{_format_type(args[0])}, {_format_type(args[1])}]"
+        elif str(origin).__contains__("Union") or str(origin).__contains__("Optional"):
+            return " or ".join(_format_type(arg) for arg in args)
+        elif origin is None:
+            return "null"
+
+    if hasattr(type_hint, "__name__"):
+        return type_hint.__name__
+
+    return str(type_hint)
+
+
+def _unroll_types(tp) -> list[type]:
+    """
+    Unrolls typing.Union and typing.Optional types into a flat list of types.
+    """
+    origin = get_origin(tp)
+    if origin is Union:
+        # Recursively unroll each type inside the Union
+        types = []
+        for arg in get_args(tp):
+            types.extend(_unroll_types(arg))
+        # Remove duplicates while preserving order
+        return list(dict.fromkeys(types))
+    else:
+        # If it's not a Union, just return it as a single-element list
+        return [tp]
+
+
+def generate_model_doc(model: type[BaseModel]) -> str:
+    """Generate documentation for a Pydantic model."""
+
+    models_stack = [model]
+
+    doc = ""
+    while models_stack:
+        current_model = models_stack.pop()
+
+        doc += f"<h4>{current_model.__name__}</h4>\n"
+
+        doc += "\n| Field Name | Type | Description |\n"
+        doc += "|------------|------|-------------|\n"
+
+        base_models = []
+        if hasattr(current_model, "__mro__"):
+            base_models = current_model.__mro__
+        else:
+            base_models = [current_model]
+
+        for base_model in base_models:
+            # Check if this is a Pydantic model
+            if hasattr(base_model, "model_fields"):
+                # Iterate through fields of this model
+                for field_name, field in base_model.model_fields.items():
+                    # Extract description from Annotated field if possible
+                    description = field.description or "No description provided."
+                    description = format_allowed_values_description(description)
+                    description = format_variable_names(description)
+
+                    # Handle Annotated types
+                    original_type = field.annotation
+                    if get_origin(original_type) is Annotated:
+                        # Extract base type and additional metadata
+                        type_args = get_args(original_type)
+                        base_type = type_args[0]
+                    else:
+                        base_type = original_type
+
+                    field_type = _format_type(base_type)
+                    field_type = format_variable_names(field_type)
+
+                    doc += f"| `{field_name}` | {field_type} | {description} |\n"
+
+                    for field_type in _unroll_types(base_type):
+                        if issubclass(field_type, BaseModel):
+                            models_stack.append(field_type)
+
+                # stop iterating the base classes
+                break
+
+        doc += "\n"
+    return doc
+
+
+def update_documentation():
+    """Update the documentation file with model information."""
+    doc_request = generate_model_doc(ConvertDocumentsRequestOptions)
+
+    with open(DOCS_FILE) as f:
+        content = f.readlines()
+
+    # Prepare to update the content
+    new_content = []
+    in_cp_section = False
+
+    for line in content:
+        if line.startswith("<!-- begin: parameters-docs -->"):
+            in_cp_section = True
+            new_content.append(line)
+            new_content.append(doc_request)
+            continue
+
+        if in_cp_section and line.strip() == "<!-- end: parameters-docs -->":
+            in_cp_section = False
+
+        if not in_cp_section:
+            new_content.append(line)
+
+    # Only write to the file if new_content is different from content
+    if "".join(new_content) != "".join(content):
+        with open(DOCS_FILE, "w") as f:
+            f.writelines(new_content)
+        print(f"Documentation updated in {DOCS_FILE}")
+    else:
+        print("No changes detected. Documentation file remains unchanged.")
+
+
+if __name__ == "__main__":
+    update_documentation()
--- a/tests/test_1-url-async-ws.py
+++ b/tests/test_1-url-async-ws.py
@@ -69,3 +69,9 @@ async def test_convert_url(async_client: httpx.AsyncClient):
    with connect(uri) as websocket:
        for message in websocket:
            print(message)
+
+    result_resp = await async_client.get(f"{base_url}/result/{task['task_id']}")
+    assert result_resp.status_code == 200, "Response should be 200 OK"
+    result = result_resp.json()
+    print(f"{result['processing_time']=}")
+    assert result["processing_time"] > 1.0
--- a/tests/test_fastapi_endpoints.py
+++ b/tests/test_fastapi_endpoints.py
@@ -54,6 +54,14 @@ async def test_health(client: AsyncClient):
    assert response.json() == {"status": "ok"}


+@pytest.mark.asyncio
+async def test_openapijson(client: AsyncClient):
+    response = await client.get("/openapi.json")
+    assert response.status_code == 200
+    schema = response.json()
+    assert "openapi" in schema
+
+
@pytest.mark.asyncio
 async def test_convert_file(client: AsyncClient, auth_headers: dict):
    """Test convert single file to all outputs"""
--- a/uv.lock
+++ b/uv.lock
Author	SHA1	Message	Date
DKL	025c4c8942	Cleanup. Text formatting. Fallback picture annotation.	2025-11-24 15:17:39 +01:00
DKL	8d5892b176	Revamp UI to SSR. Signed-off-by: DKL <dkl@zurich.ibm.com>	2025-11-21 16:15:36 +01:00
Michele Dolfi	e437e830c9	fix: Dependencies updates – Docling 2.63.0 (#443 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-11-21 10:31:56 +01:00
Michele Dolfi	2c23f65507	feat: version endpoint (#442 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-11-20 17:57:10 +01:00
Burt Holzman	5dc942f25b	chore: docs typo (cude -> cuda) (#437 ) Signed-off-by: Burt Holzman <burt@fnal.gov>	2025-11-17 08:31:44 +01:00
github-actions[bot]	ff310f2b13	chore: bump version to 1.8.0 [skip ci]	2025-10-31 17:01:56 +00:00
Michele Dolfi	bf132a3c3e	feat: Docling with new standard pipeline with threading (#428 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-31 17:57:38 +01:00
Michele Dolfi	35319b0da7	docs: Expand automatic docs to nested objects. More complete usage docs. (#426 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-31 15:02:20 +01:00
Michele Dolfi	f3957aeb57	docs: add docs for docling parameters like performance and debug (#424 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-31 14:17:31 +01:00
github-actions[bot]	1ec44220f5	chore: bump version to 1.7.2 [skip ci]	2025-10-30 15:14:17 +00:00
Michele Dolfi	e9b41406c4	fix: Update locked dependencies. Docling fixes, Expose temperature parameter for vlm models (#423 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-30 16:09:21 +01:00
Michele Dolfi	a2e68d39ae	test: check that processing time is not skipped (#416 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-27 08:29:05 +01:00
Michele Dolfi	7bf2e7b366	fix: temporary constrain fastapi version (#418 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-24 11:22:05 +02:00
github-actions[bot]	462ceff9d1	chore: bump version to 1.7.1 [skip ci]	2025-10-22 14:01:58 +00:00
Michele Dolfi	97613a1974	fix: Upgrade dependencies (#417 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-22 15:42:59 +02:00
Paweł Rein	0961f2c574	fix: makes task status shared across multiple instances in RQ mode, resolves #378 (#415 ) Signed-off-by: Pawel Rein <pawel.rein@prezi.com>	2025-10-21 15:08:42 +02:00
Tiago Santana	9672f310b1	docs: Generate usage.md automatically (#340 ) Signed-off-by: Tiago Santana <54704492+SantanaTiago@users.noreply.github.com> Signed-off-by: Michele Dolfi <dol@zurich.ibm.com> Co-authored-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-21 14:27:01 +02:00
Michele Dolfi	56e8535a7a	chore: publish release notes on Discord (#409 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-20 14:15:58 +02:00
Michele Dolfi	0f274ab135	fix: `DOCLING_SERVE_SYNC_POLL_INTERVAL` controls the synchronous polling time (#413 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-20 14:14:00 +02:00
Michele Dolfi	0427f71ef4	chore: allow to change the container runtime (#412 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-20 14:13:51 +02:00
github-actions[bot]	b6eece7ef0	chore: bump version to 1.7.0 [skip ci]	2025-10-17 12:16:37 +00:00
Michele Dolfi	f5af71e8f6	feat(UI): add auto and orcmac options in demo UI (#408 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-17 12:23:57 +02:00
Michele Dolfi	d95ea94087	feat: Docling with auto-ocr (#403 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-15 21:15:29 +02:00
sahlex	5344505718	fix: run docling ui behind a reverse proxy using a context path (#396 ) Signed-off-by: Sahler.Alexander <Alexander.Sahler@m-net.de> Signed-off-by: sahlex <1122279+sahlex@users.noreply.github.com> Co-authored-by: Sahler.Alexander <Alexander.Sahler@m-net.de>	2025-10-09 16:07:02 +02:00
github-actions[bot]	5edc624fbf	chore: bump version to 1.6.0 [skip ci]	2025-10-03 13:39:59 +00:00
Michele Dolfi	45f0f3c8f9	fix: update locked dependencies (#392 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-03 15:33:45 +02:00
Michele Dolfi	0595d31d5b	feat: pin new version of jobkit with granite-docling and connectors (#391 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-03 14:24:51 +02:00
Michele Dolfi	f6b5f0e063	docs: fix docs for websocket breaking condition (#390 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-10-02 10:55:00 +02:00
Michele Dolfi	8b22a39141	fix(UI): allow both lowercase and uppercase extensions (#386 ) Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>	2025-09-29 09:40:49 +02:00
erikmargaronis	d4eac053f9	fix: Correctly raise HTTPException for Gateway Timeout (#382 ) Signed-off-by: Erik Margaronis <erik.margaronis@gmail.com>	2025-09-29 08:06:21 +02:00
Rui Dias Gomes	fa1c5f04f3	ci: improve caching steps (#371 ) Signed-off-by: rmdg88 <rmdg88@gmail.com>	2025-09-23 18:15:12 +02:00
Viktor Kuropiatnyk	ba61af2359	fix: Pinning of higher version of dependencies to fix potential security issues (#363 ) Signed-off-by: Viktor Kuropiatnyk <vku@zurich.ibm.com>	2025-09-18 08:57:41 +02:00