Vector search that fits in a SQLite file

What it does

sqlite-vec is a C extension that adds vector search to SQLite through a vec0 virtual table. You store float, int8, or binary vectors alongside your regular data and run k-nearest-neighbor queries with ordinary SQL—no external vector database required.

The interesting bit

The author calls it “extremely small, ‘fast enough’"—a refreshing admission in a field obsessed with benchmark warfare. Pure C, zero dependencies, and it compiles to WASM for browser use. The same .db file that holds your users and orders now holds your embeddings too.

Key highlights

• Supports float, int8, and binary vector types
• Bindings for Python, Node.js, Ruby, Go, Rust, Datasette, rqlite, and sqlite-utils
• Metadata and auxiliary columns keep vector + relational data in one table
• Backed by Mozilla Builders; also sponsored by Fly.io, Turso, SQLite Cloud, and Shinkai

Caveats

• Pre-v1, so breaking changes are expected
• The README doesn’t quantify “fast enough”—no throughput or latency numbers given

Verdict

Worth a look if you’re building local-first or edge AI apps and resent maintaining a second database just for embeddings. Skip it if you need proven scale guarantees or are allergic to pre-1.0 software.