Kuzu V0 120 — Better

(If you’re still on the older 0.11.x line, buckle up – the new 0.12.0 release brings enough goodies to make the upgrade feel like moving from a bike to a sports car.)

Kùzu v0.12.0 made major strides in its "Zero-Dependency" philosophy:

If you were looking for a specific bug fix or feature introduced in a specific version not covered here, please clarify the context

[Unstructured Data] ──> [LLM Entities] ──> [Kùzu v0.12.0 Graph] ──> [Contextual Search] Native Vector & Full-Text Bundling

: Users can index text properties directly, allowing for high-performance keyword searches within the graph PyPI - kuzu . 3. Better Scalability: Out-of-Memory Performance kuzu v0 120 better

LOAD FROM "logs.parquet" WHERE timestamp > "2023-01-01" RETURN count(*);

Traditionally, graph database management systems (GDBMSs) have been built as standalone, distributed servers. While effective for massive transactional environments, this setup creates massive latency bottlenecks during analytical workloads due to network overhead, serialization costs, and rigid server management.

[Actual] Q: your question A:

Data architectures are shifting rapidly toward decentralized, lightweight, and single-node processing frameworks. In the relational database world, DuckDB proved that an in-process, serverless engine could outperform massive distributed data warehouses for analytical workloads. In the graph computing space, is pulling off the exact same feat. (If you’re still on the older 0

Adjacency lists are stored using a highly compressed, Columnar Sparse Row (CSR) matrix design. This structure allows Kuzu v0.12.0 to perform extremely fast index-free adjacency lookups. Traversing an edge requires zero traditional B-tree index lookups—it simply computes an array offset in memory, resulting in multi-fold performance gains for dense networks. 3. Native Vector and Full-Text Search Indices

| Scenario | Recommended Mode | |----------|-------------------| | | In‑memory (default) | | Cold historical data | On‑disk columnar files (compressed with LZ4) | | Mixed workloads | Hybrid (hot vertices in RAM, edges on SSD) |

We are thrilled to announce the release of !

Client-Server Architecture (Legacy) [Application] <--- Network (TCP/Bolt) ---> [Graph Database Server] Embedded Architecture (Kùzu) [Application Library <--- Shared Memory ---> Kùzu Core Engine] The In-Process Edge In the graph computing space, is pulling off

: This is Kùzu's "secret sauce." It avoids the exponential growth of intermediate results during complex joins (a common problem in graph databases), making it better at handling multi-hop queries that would crash traditional systems CIDR 2023 - KŮZU . 2. Modern Graph Features: Vector Indices & Full-Text Search

All tests ran on a 32‑core Intel Xeon 2.6 GHz with 256 GiB RAM and a 2 TB NVMe SSD.

Kuzu v0.12.0 isn't just incrementally faster; its fundamental architecture makes it structurally superior for multi-hop graph queries. 1. Vectorized and Factorized Execution