LightRAG Exploration

LightRAG Exploration is a Python FastAPI application for exploring Retrieval-Augmented Generation (RAG) workflows. It provides a modern web interface for chat, graph visualization, and RAG experimentation using local LLMs via Ollama. The app is designed for easy local use and sharing as a pipx CLI tool.

Features

• RAG Chat: Interact with a RAG pipeline using your own documents and local LLMs.
• Classic Chat: Simple chat interface with model selection and context summarization.
• Graph Viewer: Visualize the entity/relation graph extracted from your research data.
• Modern Web UI: Clean, responsive interface built with Tailwind CSS.
• Easy Setup: Installable as a pipx CLI for global use.
• Ollama Integration: Uses Ollama for local LLM inference and embeddings.
• Configurable: Uses .env for research output directory and Ollama host.

Prerequisites

• Python 3.9+
• pipx
• Ollama installed and running with required models (e.g. gemma3:1b, nomic-embed-text:latest)

Install system dependencies (macOS example)

brew install pipx
pipx ensurepath
source ~/.zshrc  # or restart your terminal

Install with pipx

pipx install lightrag-exploration

Usage

• The API will run on http://localhost:8080 by default.
• Open your browser and navigate to:
  • /frontend — Home page
  • /frontend/chat — Classic chat interface
  • /frontend/rag-chat — RAG chat interface
  • /frontend/graph-viewer — Graph visualization
  • /docs — FastAPI API docs

How It Works

• RAG Pipeline: Uses LightRAG and Ollama for retrieval, embedding, and LLM completion.
• Templates: Web UI is served from the templates/ directory.
• Config: Set research_output_dir and ollama_host in .env or .env.example.

Development

• All source code is in the root and routers/ directory.
• The entrypoint is main.py, which runs the FastAPI app.
• You can run locally with:

  python main.py
  

• Or install and run globally with pipx as above.

Test your pipx build locally before uploading to PyPI!

You can install your built wheel or sdist with pipx to verify everything works as expected:

python -m build
pipx install --force dist/lightrag_exploration-0.1.0-py3-none-any.whl
# or
pipx install --force dist/lightrag_exploration-0.1.0.tar.gz
lightrag-exploration
pipx uninstall lightrag-exploration

Packaging & Distribution

To build and publish to PyPI:

1. Upgrade build and twine:

   python -m pip install --upgrade build twine
   

2. Build your package:

   python -m build
   

3. (Optional) Check your package:

   python -m twine check dist/*
   

4. Upload to PyPI:

   python -m twine upload dist/*
   

Troubleshooting

• If you see errors about missing templates, ensure you installed with pipx after running pipx ensurepath and that your terminal session is up to date.
• Make sure Ollama is running and the required models are pulled.
• If you see errors about missing research output, check your .env and that the directory exists.

License

MIT