中文 | English

A database-native Agent Harness by the OceanBase OSS Team.

AgentSeek helps teams turn agent runtime data into a database workload: turns, context, tool calls, tasks, feedback, checkpoints, memory, and observability stay queryable instead of being scattered across logs and side systems.

It is built for teams that want to move from a local LangChain, DeepAgents, or Bub prototype to a maintainable agent application with a clear runtime, storage, context, and serving story.

"Deep Agents 实战": a free LangChain / DeepAgents course with AgentSeek labs. Course site · Source repo

uv tool install agentseek
agentseek create --list-templates
agentseek create langchain/markdown-messages
cd markdown_messages_agent
cp .env.example .env
uv sync
uv run langgraph dev

Use this path when you want a generated application scaffold. Templates cover LangChain, DeepAgents, and lightweight starters.

agentseek chat

Use this path when you want the harness runtime directly: a chat loop, gateway, plugins, MCP, or an embeddable Python package.

For all commands and options, see the CLI reference.

Related projects:

AgentSeek also builds on Bub, a hook-first agent runtime and framework.

The usual flow is:

1. Create a project when you need an application scaffold.
2. Run AgentSeek itself when you need the harness runtime.
3. Add durable runtime data through the harness and storage integrations.
4. Add semantic memory with ContextSeek when the agent needs cross-session context.
5. Serve production LangGraph apps through agentseek-api.

Storage, memory, and serving are added only when the application needs them.

Pick the smallest template that matches your application:

See Templates reference for the full catalogue.

• Documentation home
• Tutorials
• How-to guides
• Explanation
• Reference

Useful package docs in this repo:

• agentseek-langchain
• agentseek-tapestore-oceanbase
• agentseek-contextseek
• agentseek-schedule-sqlalchemy

make install
make check
make test
make docs-test

Apache-2.0