SeekDB: A MySQL-Compatible State Store for AI Agents

If you're building AI agents that need persistent, queryable state—and you don't want to abandon SQL—SeekDB is worth a look. It’s a lightweight, MySQL-compatible database purpose-built for agent workflows. Think of it as a state store that speaks SQL, so you can search across agent memories, forks, and histories without learning a new query language.

What It Does

SeekDB is a database engine that stores and queries agent state (conversations, tool outputs, knowledge graphs, etc.) using MySQL syntax. You can INSERT new observations, SELECT across agent runs, and even fork a session’s state to experiment with alternative decisions. It’s designed to fit into agent frameworks like LangChain, AutoGPT, or custom pipelines.

The key idea: agents produce a lot of ephemeral data. Most teams store that in JSON blobs or vector stores, but lose the ability to do relational queries. SeekDB gives you both—structured storage plus agent-friendly features like state branching.

Why It’s Cool

A few features stand out:

• MySQL compatibility – You don’t need a new driver or DSL. If you know SELECT, INSERT, and CREATE TABLE, you can use it. That lowers the barrier for onboarding.

• State forking – During an agent run, you can fork the entire state (like a Git branch). This is huge for debugging or A/B testing decisions. Want to see what happens if the agent chooses tool A instead of tool B? Fork, change the action, and replay.

• Search over agent history – Because state is stored in tables, you can query across thousands of agent conversations, filter by timestamp, action type, or any custom field. No need to dump JSON and grep.

• Lightweight – It’s a single binary or Docker image. No heavy infrastructure required. Perfect for local development or edge deployments.

How to Try It

The easiest way is via Docker:

docker run -p 3306:3306 oceanbase/seekdb:latest

Then connect with any MySQL client:

mysql -h 127.0.0.1 -P 3306 -u root

From there, you can create tables for agent state, insert observations, and query them. The repo has a quickstart guide and example schema for a simple chat agent.

If you prefer building from source, clone the repo and follow the build instructions in the README.

Final Thoughts

SeekDB is solving a real pain point: agent state management. Most teams either hack something with a vector DB or use Redis with limited queryability. This gives you the flexibility of SQL plus agent-specific features like forking. It’s still early stage, but the design is clean and the MySQL interface makes it instantly familiar.

If you’re prototyping agents or running production workloads that need persistent, searchable state, give SeekDB a spin. It might save you from writing yet another state management shim.

Found this useful? Follow us at @githubprojects for more developer tools and open source highlights.