Back to directory

ferrovec

An in-browser vector store built as a Rust-to-WebAssembly HNSW engine that persists to the browser's private file system and stays consistent across tabs.

ferrovec

Most in-browser vector libraries keep an in-memory index that disappears on reload and can diverge once a second tab opens. ferrovec instead stores its HNSW graph on disk in the browser's Origin Private File System and uses a single-writer leader election over Web Locks and BroadcastChannel so multiple open tabs share one consistent store instead of silently diverging, rehydrating automatically when the store is reopened.

It supports incremental upsert-style inserts, tombstoned removals, and in-place compaction, so adding one vector does not require rebuilding the whole index. The core is a hand-rolled HNSW implementation in Rust with only serde and postcard as dependencies, compiling to roughly 33KB gzipped with a deterministic pseudo-random generator, no threads, and no unsafe code outside one audited kernel, plus a SIMD-accelerated distance path with a scalar fallback.

The npm package adds transformers.js auto-embedding and a three-line JavaScript API on top of the Rust and WebAssembly core, which is also published separately to Rust's crates.io registry.

More in Databases

See all 294
  • SQLBraid

    clickin.github.io

    A SQL-first data-access toolkit for TypeScript that keeps ordinary SQL visible in tagged template literals, binds interpolated values safely by default, supports readable conditional SQL directives, and lets a schema validate the returned rows.

  • A SQL-style query language and toolchain for the Qdrant vector database, covering dense, sparse and hybrid search, filtering and multitenancy, with the same grammar running from a CLI, Python, Node.js, Rust, WebAssembly and an in-process edge engine.

  • Fast Knowledge

    github.com

    A self-hosted, Java-based knowledge base for small and mid-sized organizations, turning internal documents into a searchable, question-answerable and audit-exportable knowledge asset with hybrid retrieval and Docker deployment.