Kuzu is an embedded, high‑performance graph database that runs directly inside applications without requiring external servers, offering both on‑disk and in‑memory storage. It provides a columnar, vectorized execution engine with novel join algorithms, supports the Cypher query language, and includes built‑in full‑text and HNSW vector indexes for AI‑driven knowledge graphs. The open‑source, MIT‑licensed platform integrates with popular data and AI ecosystems such as LangChain, PyTorch Geometric, and Pandas.
Funding
Funding not disclosed
Founders
Product
Problem
Developers and data teams often need to query complex relationships and perform vector search on knowledge graphs, but existing graph databases either require separate server processes, lack high-performance execution, or do not integrate smoothly with AI and data tooling.
Solution
Kuzu offers an embedded, high‑performance graph database that runs directly inside the host application, eliminating the need for external servers. Its columnar storage, vectorized query engine, and novel worst‑case‑optimal join algorithms enable fast, scalable execution of Cypher queries on both on‑disk and in‑memory graphs. Built‑in full‑text and HNSW vector indexes support AI‑driven knowledge‑graph and vector‑search workloads. The MIT‑licensed engine integrates natively with popular AI and data ecosystems such as LangChain, PyTorch Geometric, LlamaIndex, Pandas, and data lake formats, allowing developers to build real‑time knowledge bases and graph‑AI applications with minimal friction.
Target Audience
Primary users are developers, data engineers, and AI teams building applications that require high‑performance graph queries, knowledge‑graph construction, or vector search, especially those who prefer an embedded database model.
Features
- In‑process architecture that embeds the database within the application, removing external server dependencies
- Columnar storage and vectorized processing for high throughput on large property graphs
- Novel worst‑case‑optimal join algorithms and morsel‑driven parallelism for fast query execution
- Support for the Cypher query language with full‑text search and HNSW vector indexes
- Compatibility with on‑disk and in‑memory storage modes
- Native bindings for Python, Node.js, Rust, Go, Java, Swift, and WebAssembly
- Seamless integration with AI and data tools (LangChain, PyTorch Geometric, LlamaIndex, Pandas, Parquet, Iceberg)