turbo-quant

Cognitive memory backbone for AI agents. Biologically-inspired 4-tier memory, it remembers, forgets, consolidates, and forms associations across a memory graph — Hebbian co-activation, temporal chains, and entity links.

A clean-room implementation of Google Research's TurboQuant, with the RaBitQ unbiased-estimator correction. Runs anywhere JavaScript runs; Node, browsers, Bun, Cloudflare Workers, React Native.

SIMD-accelerated 4/8-bit vector quantization for approximate nearest neighbor search, based on TurboQuant (ICLR 2026). Standalone C++17 library with Python bindings

Lightweight, fast, secure, and free document chat system powered by Qwen AI and TurboVec search.

Gemma-4-E4B-it running locally in a browser with WebGPU on a M2 Mac.