Docling Ingest

A web application for PDF document ingestion with interactive preview, content editing, and vector database storage. Built on Docling for PDF parsing and ChromaDB for vector storage.

Features

• PDF Upload & Conversion — Upload PDFs, automatically extract text, tables, and images using Docling
• Interactive Content Viewer — Preview extracted content page-by-page with visual annotations
• Content Editing — Delete, reorder, and edit items before ingestion. Full undo/redo support
• Image Classification — Classify images (logo, chart, diagram, photo, etc.) with customizable presets
• AI Image Descriptions — Optional vision model integration to auto-generate image alt-text
• Embedding Preview — Preview how documents will be chunked before ingestion
• Vector DB Ingestion — Ingest processed documents into ChromaDB collections
• Multiple Strategies — Choose between per-page, chunked, or custom ingestion strategies
• Dark Mode — Automatic theme detection based on system preferences
• Offline-First — Document state is managed in the browser via localStorage

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────────┐
│   Frontend   │────▶│    Proxy     │────▶│     Engine       │
│  React/Vite  │     │   Node.js   │     │  Python/FastAPI  │
│   :3000      │     │   :4006     │     │   :8000          │
└─────────────┘     └─────────────┘     └─────────────────┘
                                              │
                                         ┌────┴────┐
                                         │ChromaDB │
                                         │(embedded)│
                                         └─────────┘

• Frontend — React + Vite + Tailwind CSS. Single-page app with the ingestion interface
• Proxy — Node.js/Express. Aggregates local and remote config, proxies requests to the engine
• Engine — Python/FastAPI. Runs Docling for PDF conversion, manages embeddings and ChromaDB

Quick Start

Docker Compose (Recommended)

git clone https://github.com/liyanfeng129/docling-ingest.git
cd docling-ingest
docker compose up --build

Open http://localhost:3000 in your browser.

With Vision Model (Optional)

To enable AI-powered image descriptions:

docker compose --profile vision up --build

Then pull the vision model:

docker exec -it docling-ingest-ollama-1 ollama pull granite3.2-vision:2b

And set ENABLE_VISION_MODEL=true in the engine environment.

Manual Setup

Engine (Python):

cd backend/engine
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --host 0.0.0.0 --port 8000

Proxy (Node.js):

cd backend/proxy
npm install
PORT=4006 ENGINE_URL=http://localhost:8000 node index.js

Frontend:

cd frontend
npm install
npm run dev

Configuration

Environment Variables

Variable	Service	Default	Description
VITE_INGESTION_URL	Frontend	http://localhost:4006	Proxy service URL
PORT	Proxy	4006	Proxy listen port
ENGINE_URL	Proxy	http://localhost:8000	Engine service URL
ENABLE_VISION_MODEL	Engine	false	Enable Ollama vision model
EMBEDDING_MODEL	Engine	Snowflake/snowflake-arctic-embed-l	Sentence Transformers model
CHROMA_PERSIST_DIRECTORY	Engine	./resources/chroma_db/default	ChromaDB storage path
CHROMA_COLLECTION_NAME	Engine	documents	Default collection name
OLLAMA_BASE_URL	Engine	http://localhost:11434	Ollama server URL
VISION_MODEL	Engine	granite3.2-vision:2b	Ollama vision model name

Persistent Data & Docker Volumes

When running with Docker Compose, any files created by the app at runtime (such as the ChromaDB vector database) are stored in named Docker volumes, not on your local filesystem directly.

Where is the Vector DB?

When you ingest a document, ChromaDB writes its data to the chroma_data volume, mapped to /app/resources/chroma_db inside the engine container.

To browse the data in Docker Desktop:

1. Open Docker Desktop
2. Click Volumes in the left sidebar
3. Click docling-ingest_chroma_data
4. Click the Data tab to browse the files

The ChromaDB collection files will be under the default/ folder.

To inspect via terminal:

# List files inside the volume
docker exec -it docling-ingest-engine-1 ls /app/resources/chroma_db/default

# Copy the entire database to your local machine
docker cp docling-ingest-engine-1:/app/resources/chroma_db ./chroma_db_backup

To use a local folder instead of a Docker volume (so data appears directly in your project), replace the volume in docker-compose.yml:

volumes:
  - ./chroma_db:/app/resources/chroma_db   # bind mount — visible in Finder/Explorer

Note: The chroma_data volume persists across container restarts and rebuilds. To fully delete it run docker volume rm docling-ingest_chroma_data.

License

MIT