A WIP CLI tool for querying technical documentation using Perplexity AI's Sonar LLM with RAG (Retrieval Augmented Generation) capabilities. I'm mostly using this to learn and experiment with different LLM and RAG approaches as well as building around them. Expect the code to be a better reference than a functional tool, at least in its current state.
- Query documentation using natural language
- RAG support using DuckDB and vector similarity search
- Support for Markdown, RST, and text files
- Code example extraction and highlighting
- Source citation tracking
- Optional in-memory or persistent vector database
- GPU acceleration support (CUDA and MPS)
src/refassist/ml/notebooks/ contains some notebooks I've used to experiment with different RAG approaches.
Requires Python 3.12.8 or higher.
git clone https://github.com/yourusername/refassist.git
cd refassist
uv sync --all-extras --devPERPLEXITY_API_KEY=your_api_key_hereBasic usage:
python -m refassist.main --file /path/to/docsWith RAG enabled:
python -m refassist.main --file /path/to/docs --store-filesOptions:
--file: Path to documentation file or directory--api-key: Your Perplexity AI API key (optional if set in .env or $PERPLEXITY_API_KEY is set)--store-files: Enable RAG processing and storage--in-memory: Use ephemeral in-memory database--db-path: Custom path for persistent database storage
- Install development dependencies:
uv sync --all-extras --dev- Run tests:
pytest tests --cov=src- Run pre-commit hooks:
pre-commit run --all-filesBuild and run with Docker:
docker build -t refassist .
docker run -it --env-file .env refassist