codemem-vector

HNSW vector index using usearch for 768-dimensional cosine similarity search.

Overview

Implements the VectorBackend trait from codemem-core. Provides fast approximate nearest neighbor search for memory and code symbol embeddings.

Configuration

Key Operations

• insert(id, embedding) — Add a vector to the index
• search(query, k) — Find k nearest neighbors
• remove(id) — Remove a vector
• save(path) / load(path) — Persist to disk at ~/.codemem/codemem.idx