LEANN Skill

LEANN (Learned Embedding ANchor Navigation) - A graph-based selective recomputation system achieving 97% storage reduction for local RAG indexes while maintaining fast retrieval performance.

Core Principle

Use leann for persistent local code indexes with minimal storage overhead.

Instead of storing all embeddings at full precision (3 GB for 1M files), leann stores only strategic anchors and reconstructs others on-demand (128 MB for same dataset - 95.8% reduction).

When to Use LEANN

Primary Use Cases