sqlite-vec
Zero-dependency SQLite extension for vector search that runs anywhere SQLite does, even WASM.
About
A vector database in a single C file, sqlite-vec is an SQLite extension by Alex Garcia that stores and queries float, int8, and binary vectors in vec0 virtual tables using plain SQL. It has zero dependencies, so it runs anywhere SQLite runs: Linux, macOS, Windows, Raspberry Pi class devices, and browsers through WASM. Bindings exist for Python, JavaScript, Ruby, Go, and Rust, alongside Datasette, sqlite-utils, and rqlite integrations, and installation is as simple as pip install sqlite-vec or the npm, gem, and cargo equivalents. Queries currently run exhaustive scans rather than approximate indexes, fast enough for the small-to-medium collections the tool targets, with int8 and binary vector types available to shrink storage; the project is pre-v1, so breaking changes are possible. Development is backed by the Mozilla Builders program with sponsorship from Fly.io, Turso, and SQLite Cloud, and both Turso and SQLite Cloud ship it in their products as the successor to sqlite-vss. Dual licensed MIT or Apache-2.0, it has around 7,900 GitHub stars.
Should you use sqlite-vec?
Pick it when
Pick sqlite-vec when you already use SQLite or need vector search on a phone, a Raspberry Pi class device, or in the browser via WASM, with small to medium collections queried through plain SQL and zero extra dependencies.
Look elsewhere when
Skip it for large collections: queries run exhaustive scans with no approximate index, and the project is pre-v1 with possible breaking changes. LanceDB stays embedded but handles billions of vectors, and pgvector suits server-side SQL.
Alternatives to sqlite-vec
- LanceDB
Also embedded with no server, but with ANN indexes, full text search, and versioning for far larger data; heavier than a single C file.
- pgvector
SQL vector search with HNSW and IVFFlat indexes in Postgres; needs a database server, but scales well beyond exhaustive scans.
- USearch
Embeddable HNSW engine with WASM and mobile bindings for approximate search at larger sizes; you manage metadata outside of SQL.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- Vector Databases & Embeddings
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Easy (2/5)
- License
- MIT OR Apache-2.0
- Added
- Jul 29, 2026