Kuzu

Kùzu is an embedded, open-source (MIT) graph database designed to be the graph equivalent of SQLite or DuckDB: it runs in-process inside your application with no server to manage, speaks Cypher, and stores labeled property graphs on disk with columnar storage and multi-core parallelism. It is a natural fit for knowledge graphs, Graph RAG backends, recommendations, fraud detection, and any workload where relationships matter more than rows.

It integrates with the Python data stack (pandas import/export, NetworkX, PyTorch Geometric), handles larger-than-memory graphs via disk-backed traversals, and interoperates with DuckDB through ATTACH so relational and graph data can be queried together. Community projects increasingly pair it with DuckDB for agentic memory architectures.

Honest limitations: as an embedded engine it is not a managed graph service, so clustering, replication, and multi-writer production topologies are your problem. The graph-database ecosystem is smaller than the relational one, meaning fewer hosted options and less Stack Overflow archaeology when something breaks.

pip install kuzu (or the Rust/JS bindings) and start querying with Cypher; no server setup required.