This enables much tighter integration with vector databases and AI frameworks like PyG or DGL. It allows for more efficient Retrieval Augmented Generation (RAG) workflows where semantic similarity is filtered by graph structure, enabling more accurate and context-aware results. 3. Expanded Ecosystem Support: Azure and Swift
To cut down on boilerplate orchestration code, version 0.12.0 expands its ecosystem with a native . This module allows Kuzu to communicate directly with external provider APIs (such as OpenAI, Anthropic, or Hugging Face) to generate text embeddings directly during data ingestion. This limits the reliance on complex, multi-tool data transformation pipelines like LangChain or LlamaIndex for basic embedding generation.
[Your Response]
and LlamaIndex for immediate ingestion of agentic memory and structured retrieval chains. kuzu v0 120 better
This query counts rows in a Parquet file without ever loading the data into the Kuzu database files, providing a zero-copy analysis experience.
: Kuzu's support for Cypher and the Bolt protocol ensures a high degree of compatibility with existing Neo4j applications and tools, reducing the barrier to entry for new users.
Handling dates and timestamps is now smoother. We have improved parsing logic for standard ISO 8601 formats and added support for duration() arithmetic. This enables much tighter integration with vector databases
import kuzu # Initialize an on-disk database and connection db = kuzu.Database("./knowledge_graph") conn = kuzu.Connection(db) # Create a simple schema conn.execute("CREATE NODE TABLE User(id INT64, name STRING, PRIMARY KEY(id))") conn.execute("CREATE NODE TABLE Skill(id INT64, name STRING, PRIMARY KEY(id))") conn.execute("CREATE REL TABLE KNOWS(FROM User TO User)") conn.execute("CREATE REL TABLE HAS_SKILL(FROM User TO Skill)") # Fast data insertion using Cypher conn.execute("CREATE (:User id: 1, name: 'Alice')") conn.execute("CREATE (:User id: 2, name: 'Bob')") conn.execute("CREATE (:Skill id: 101, name: 'GraphRAG')") conn.execute("CREATE (:User id: 1, name: 'Alice')-[:KNOWS]->(:User id: 2, name: 'Bob')") conn.execute("CREATE (:User id: 1, name: 'Alice')-[:HAS_SKILL]->(:Skill id: 101, name: 'GraphRAG')") # Query the graph response = conn.execute( "MATCH (a:User)-[:KNOWS]->(b:User)-[:HAS_SKILL]->(s:Skill) " "RETURN a.name, b.name, s.name" ) while response.has_next(): print(response.get_next()) Use code with caution. 🎯 Ideal Use Cases for v0.12.0 Releases · kuzudb/kuzu - GitHub
Rather than clogging the core binary with unneeded features, Kùzu isolates specialized functionality into high-performance native extensions. Following the trajectory of previous versions, v0.12.0 provides seamless, out-of-the-box maturity for four primary subsystems:
: Uses a table-based storage model that allows for efficient columnar data access and vectorized query execution. Expanded Ecosystem Support: Azure and Swift To cut
and DGL to train Graph Neural Networks (GNNs) without external pipeline tools.
"I didn't believe the hype. I put the V0 120 on a 7-inch angle grinder with a flap disc configuration. I prepped a 4-foot weld seam in 3 minutes. Normally, that takes 8 minutes. It's not just better; it's a different category." —
This article dives into the key enhancements in Kùzu v0.12.0 and why it's a critical upgrade for graph developers. 1. Substantial Performance Boosts
The aspect is most visible when working with aluminum or soft brass. Traditional 120-grit wheels load up (clog) within 30 seconds. The Kuzu V0 120 features a wide, shallow chip valley that ejects swarf rather than retaining it. Users report zero loading even after 10 minutes of continuous use on 6061 aluminum.
To see the full technical details, you can visit the Kùzu GitHub repository .