HyperspaceDB
A high-performance vector database using 1-bit quantization and hyperbolic geometry to model hierarchical data, aimed at autonomous agents and robotics rather than static document search.
HyperspaceDB is a vector database that represents hierarchical data, such as codebases or taxonomies, in hyperbolic space using Poincare and Lorentz models rather than plain Euclidean vectors, which it describes as compressing large trees into lower-dimensional spaces with less memory use. It targets continuous, episodic memory for autonomous agents and robots rather than static search over fixed documents.
It uses 1-bit quantization and a lock-free, SIMD-accelerated architecture for search and ingest throughput, includes an SDK with Riemannian-geometry operations such as computing a Frechet mean or parallel transport for updating vectors over time, a Merkle-tree delta-sync protocol for asynchronously syncing edge devices with the cloud, and role-based access control for sharing data across tenants.
It is written in Rust and released under the MIT license.