V0 136: Kuzu
: Enhanced the HNSW (Hierarchical Navigable Small World) index by compressing neighbor offsets for in-memory graphs, reducing memory footprint. Expanded Data Support : Added support for columns in vector indices and improved session token parameter support for cloud-based storage. Language & API Updates TypeScript : Added formal TypeScript definitions for the database API.
No more manual installation for algo , fts , json , and vector extensions. kuzu v0 136
Data scientists building RAG systems (Retrieval-Augmented Generation) need to store entity relationships. The new LIST of STRUCT type allows you to attach vector embeddings directly to nodes as a list of floats, eliminating the need for a separate vector database. : Enhanced the HNSW (Hierarchical Navigable Small World)
: Added support for the ignore_pattern option, allowing for more flexible search indexing. No more manual installation for algo , fts
: Implementation of an 8-byte hash slot for the AggregateHashTable to improve aggregation efficiency.
From a technical standpoint, Kuzu v0.136 appears to be built using a combination of modern programming languages, including C++, Rust, and Python. The project leverages several open-source libraries and frameworks, such as the Boost C++ Libraries and the pybind11 Python binding generator.