Vector Search
On-device similarity search using cosine, dot, or euclidean metrics. Combine with metadata filters in one query — find the 5 most relevant documents for a given embedding without a cloud round-trip.

Documents and vectors, on device. The embedded database for the AI era.
On-device similarity search using cosine, dot, or euclidean metrics. Combine with metadata filters in one query — find the 5 most relevant documents for a given embedding without a cloud round-trip.
The engine is written in Rust and compiles to a sub-400 KB WASM bundle, a native .node module, and a static library for React Native — the same code on every platform.
Familiar find, insert, update, delete with $eq, $gt, $in, $and, $or and more. Fully typed with TypeScript generics — including vector index methods.
Type-safe B-tree indexes with O(log n) range scans. The query planner picks the best index automatically.
Powered by redb — a pure-Rust B-tree storage engine. Every write is atomic, consistent, isolated, and durable.
Subscribe to a collection with a filter and receive a fresh snapshot after every write — without polling.
Wrap any backend with EncryptedBackend for transparent AES-GCM-256 encryption. Keys derived via PBKDF2-HMAC-SHA256.
Browser (WASM + OPFS SharedWorker), Node.js (napi-rs), and React Native (JSI). One package, one API, zero platform branches.
