LanceDB
LanceDB is an open-source embedded retrieval library for multimodal AI, built on the Lance columnar format. Repository description: “Developer-friendly OSS embedded retrieval library for multimodal AI”.
Repository snapshot (GitHub API, 2026-11-19)
| Item | Detail |
|---|---|
| Repo | lancedb/lancedb, Rust, Apache-2.0 |
| Created | 2023-02-28 |
| Activity | last push 2026-11-19; latest release v0.41.0 (2026-10-29) |
| Size | 11,947 stars, 1,171 forks |
Capabilities (docs.lancedb.com)
- Vector indexes: IVF_PQ (general default), IVF_RQ, IVF_HNSW_SQ and IVF_FLAT (required for binary Hamming search); more IVF and HNSW variants in the embedded library. A BM25 full-text index is separate.
- Metrics: l2 (default), cosine, dot and hamming. The metric is fixed when the index is built; a different one at query time falls back to brute force.
- Hybrid search, full-text search, reranking and SQL filtering are listed. Sparse vectors were not confirmed.
- Automatic data versioning (README). IVF_RQ needs dimensions divisible by 8; no numeric dimension limit is stated.
- SDKs for Python, TypeScript and Rust, plus a REST API.
Deployment
LanceDB Embedded runs inside your application process on local disk or object storage. LanceDB Enterprise is the paid distributed service with private deployment. Its billing model was not stated on the pages read.
Where it fits
LanceDB is the in-process option for local-first apps and data science workflows, like chroma and unlike client-server qdrant or milvus. See vector-databases-compared.
Sources
- docs.lancedb.com, its indexing page, the LanceDB README and GitHub API (2026-11-19)