LEANN is a lightweight, privacy-first retrieval system that indexes and searches millions of documents while using 97% less storage than traditional solutions without losing accuracy.
It avoids storing all embeddings by using graph-based selective recomputation with high-degree preserving pruning and computes embeddings only when needed.
Key Features:
- Privacy-first: Everything runs locally, no cloud dependencies.
- Lightweight: Graph pruning + CSR format minimize storage and memory.
- Portable: Easily move knowledge bases across devices.
- Scalable: Handles messy personal and agent-generated data.
- Accurate: Matches heavyweight vector DBs while using a fraction of the storage.