CozoDB is a transactional, relational-graph-vector database designed for AI applications, offering Datalog for querying and supporting graph algorithms. It targets developers needing a versatile, embeddable database that can handle complex data relationships and vector similarity searches, providing a "hippocampus for AI."

How It Works

CozoDB leverages Datalog for its query language, enabling powerful recursive queries and composability, which is particularly advantageous for graph traversal and analysis. It integrates Hierarchical Navigable Small World (HNSW) indices for vector search directly within Datalog, allowing seamless unification with traditional relational operations. This approach allows for complex query logic, including recursive vector searches and the application of graph algorithms like community detection.

Quick Start & Requirements

• Install/Run: Try CozoDB in your browser via the WASM page: https://cozodb.github.io/cozodb-wasm/. Installation instructions for various environments (Python, NodeJS, JVM, etc.) are available: https://cozodb.github.io/cozodb/install/.
• Prerequisites: No specific prerequisites for the WASM version. Other environments may require language runtimes (Python, NodeJS, Java).
• Setup Time: Minimal for WASM. Installation into other environments depends on the chosen language binding.

Highlighted Details

• Supports HNSW vector search integrated into Datalog for complex similarity queries.
• Offers "time travel" for data, allowing historical queries on a per-relation basis.
• Embeddable across numerous platforms including mobile and web browsers (WASM).
• Provides multiple storage backends: in-memory, SQLite, RocksDB, Sled, and TiKV.

Maintenance & Community

The project is young but actively developed, with recent releases (v0.6, v0.7) introducing significant features like HNSW and MinHash-LSH. Feedback is encouraged. Project discussions are available via the issue tracker and Reddit.

Licensing & Compatibility

Licensed under MPL-2.0 or later. This license is generally permissive for commercial use and linking with closed-source applications, but requires modifications to the licensed code to be shared under the same license.

Limitations & Caveats

Versions before 1.0 do not guarantee syntax or storage compatibility. The project is described as "very young," suggesting potential for breaking changes and evolving features.