seekdb

Why we included this project

Agent teams typically end up maintaining separate stores for embeddings, full-text search, and relational rows, with hand-written code stitching the results together. seekdb replaces that tangle with a single engine that handles vector, text, structured, and semi-structured data together, so one SQL query can combine vector similarity, keyword matching, and scalar filters. It was designed as a state store for AI agents, which means conversation history, tool outputs, and long-term memory can live next to the embeddings instead of being split across systems. It runs embedded in-process or as a server, speaks MySQL protocol, and has integrations for LangChain, LlamaIndex, and Dify, so it drops into existing RAG and agent pipelines without much setup. Teams building write-then-retrieve loops with tight latency budgets, or wanting a fork/merge sandbox to test agent behavior safely, get the most from it.