LEANN
A local vector database searches personal data with compact indexes, Ollama support and optional cloud model backends under the MIT license.
LEANN is a local vector database for people building AI search over personal data or knowledge collections. It reduces index storage by computing embeddings on demand rather than keeping every embedding on disk.
- Personal data search: Search by meaning across files, email, browser history and conversations, including WeChat, iMessage, ChatGPT and Claude data.
- Visual PDF retrieval: ColQwen2 and ColPali search PDFs using text, figures and page layout, with MPS acceleration on Apple Silicon.
- Model backends: Use HuggingFace or Ollama for generation, or connect to OpenAI-compatible local servers such as LM Studio, vLLM and llama.cpp.
- Live data connections: MCP servers bring in data from services such as Slack and Twitter; those connections require service access and authentication.
Local embedding and generation backends can keep processing on your machine. LEANN also supports Anthropic and OpenAI-compatible cloud services, so where data goes depends on the providers you choose. Exported data can work offline; live service connections need internet access.
The project is open source under MIT and collects no telemetry. You can transfer indexes between devices.
Logotyper
