LEANN

This project addresses the challenge of running retrieval-augmented generation (RAG) applications locally with minimal storage overhead. Unlike typical vector databases that store embeddings for millions of documents, it uses graph-based selective recomputation and high-degree preserving pruning to compute embeddings on demand, achieving up to 97% storage savings without sacrificing accuracy. This approach suits ML engineers and AI practitioners who want private, offline-first semantic search over large personal or team data sets without relying on cloud services. With nearly 12,000 stars and recent commits, it shows strong community interest and ongoing maintenance, though its growth history is still emerging. It is not suitable if you need a mature, widely adopted vector database with extensive ecosystem integrations or cloud-native features; in such cases, alternatives like Pinecone or Weaviate would be better choices.

Use this when you need to run large-scale semantic search or RAG applications entirely on personal devices, prioritising privacy and minimal storage. If you require cloud-based scalability or a more established vector database with broader support, consider alternatives like FAISS or Pinecone instead.

Commonly deployed for personal or small-team knowledge management scenarios such as semantic search over local file systems, email archives, browser history, chat logs, or codebases. It also fits use cases involving offline agent memory or private AI assistants that index large, messy personal data sets without offloading data to the cloud.

Latest commit: 2026-05-27. Activity data is based on daily RepoPi snapshots of the GitHub repository.