yarlabs/hyperspace-db

HyperspaceDB is a high performance vector database written in Rust, aimed at AI agents, robotics, and autonomous systems. A vector database is a specialized storage system for AI embeddings, the numerical representations that AI models use to capture meaning. Unlike most vector databases designed mainly for simple document search, HyperspaceDB targets more demanding use cases involving continuous learning, spatial reasoning, and hierarchical data structures. Its standout technical feature is native support for hyperbolic geometry, a family of mathematical spaces that represent hierarchical relationships, such as organizational charts, code file trees, or biological taxonomies, more efficiently than standard flat, or Euclidean, geometry. The project claims this lets it store equivalent semantic information using far fewer dimensions, which in turn uses dramatically less memory and disk space. Other practical capabilities include one bit quantization, which compresses vector data to use eight times less storage, asynchronous replication across multiple nodes, and edge to cloud synchronization for offline first deployments, useful for robots or drones that cannot rely on a constant internet connection. It also supports multi tenant isolation for software as a service applications, and automatic offloading of rarely accessed, or cold, data to object storage such as S3. Performance benchmarks reported in the README show very high throughput compared to competing databases, with the hyperbolic geometry mode showing especially large advantages in both speed and storage. The project targets robotics teams, AI research labs building long term memory systems for agents, and enterprise applications that need to combine graph like relationships with semantic search. It is released under the AGPL version 3 license, with a separate commercial license available for organizations that need different terms.