Kuzu runs entirely in-process within your application. This design removes the overhead of network serialization, context switching, and client-server connection management. It provides direct memory access between the database and host programs written in Python, Rust, C++, Node.js, or Go. 2. Columnar Disk-Based Storage
Kùzu integrates directly with Pandas, Polars, Arrow, and NetworkX, allowing effortless data ingestion and extraction. What’s New and Improved in Kùzu v0.13.6
While Kuzu is still pre-version 1.0, it has matured rapidly. Version 0.1.36 is not a rewrite; it is a that introduces powerful usability features for developers and fixes critical edge cases in query execution. kuzu v0 136
Kùzu (pronounced “ku‑zu”) is an (GDBMS) designed for query speed and scalability. Unlike traditional client‑server databases, Kùzu is serverless and runs in‑process with your application, similar to SQLite but for graph workloads. This makes it easy to integrate into Python, Node.js, Rust, Go, Java, C/C++, and even browser‑based applications.
Kùzu v0.13.6 brings substantial under-the-hood upgrades to its cost-based query optimizer. Complex graph patterns involving multiple joins and variable-length paths now generate more predictable, high-performance execution plans. Memory allocation during large MATCH queries has also been optimized, reducing the peak memory footprint for complex analytical workloads. 2. Streamlined Vector Search & GraphRAG Workflows Kuzu runs entirely in-process within your application
While the "v0" days are behind it, getting started with the latest version of Kuzu is straightforward. You can install and use Kuzu via several client APIs:
Data scientists training Graph Neural Networks (GNNs) or calculating graph features (like PageRank, degree centrality, or shortest paths) often struggle with data movement bottlenecks. Exporting data from a centralized database across a network to a Jupyter Notebook destroys iteration speed. Version 0
April 11, 2026 Status: Unidentified / No matching reference found
Queries are executed using a vectorized execution engine. Data is processed in fixed-size blocks (vectors) rather than row-by-row. This approach maximizes CPU cache locality, minimizes instruction overhead, and allows Kùzu to leverage modern CPU features like SIMD (Single Instruction, Multiple Data). Factorized Query Execution
: Kùzu has been validated on industry-standard benchmarks like LDBC-SF100 (280 million nodes, 1.7 billion edges) and can execute complex 30-hop path queries in milliseconds on consumer hardware.
import kuzu # Initialize database on disk and connect to it db = kuzu.Database('./my_graph_db') conn = kuzu.Connection(db) # Create Node Table (Schema definition) conn.execute("CREATE NODE TABLE User(name STRING, age INT64, PRIMARY KEY (name))") # Create Relationship Table (Schema definition) conn.execute("CREATE REL TABLE Follows(FROM User TO User)") Use code with caution. Step 3: Insert Data