Embedded long-term memory runtime & vector search engine for AI agents
-
Updated
Nov 2, 2026 - Rust
Embedded long-term memory runtime & vector search engine for AI agents
Engineered a production-grade Vector Database from scratch, implementing the HNSW algorithm for high-performance semantic search with sub-millisecond latency.
HNSW (Hierarchical Navigable Small World) vector index built from scratch in Python/NumPy: no FAISS, no hnswlib. recall@10 = 1.000, 20x speedup over brute force at n=8000.
Vector Search using HNSW
Approximate nearest-neighbour search implemented from scratch in Python: brute-force kNN through to HNSW, benchmarked on SIFT1M.
To associate your repository with the hnsw-algorithm topic, visit your repo's landing page and select "manage topics."