seekdb
A MySQL-compatible embedded or server database that unifies vector, full-text, and structured search in one engine, designed as a fast, forkable state store for AI agents.
seekdb is an AI-native search database that combines vector similarity search, full-text search, and ordinary structured and semi-structured data in one MySQL-compatible engine, so an agent can run a single SQL query that filters, full-text matches, and ranks by vector distance together instead of merging results from separate systems client-side.
It supports both an embedded, in-process mode with no server or schema setup and a server or clustered mode reachable with the same client. An asynchronous index pipeline lets newly written data become searchable immediately without blocking the write, keeping latency flat under concurrent load, and a fork-and-merge mechanism gives agents copy-on-write sandboxes to experiment in and either commit or discard. It works with LangChain, LlamaIndex, and Dify.
seekdb is written in C++ and released under the Apache License 2.0.