AgentSeek¶

AgentSeek is a database-native Agent Harness for teams that want agent runtime data to become a first-class database workload. It is open to any Agent Framework — the current version ships with built-in Bub and is LangChain-friendly out of the box.

Prerequisites: Python 3.12+, uv, and a model provider API key. That's it.

AgentSeek is a suite of components that work independently or together:

Component	What it does	Docs
agentseek-cli	Scaffold projects, manage lifecycle (`create / run / build / deploy`)	ob-labs/agentseek
agentseek-api	Agent Protocol server — ship your LangGraph to production, zero code change	ob-labs/agentseek-api
ContextSeek	Semantic context layer — memory, retrieval, evolution, progressive disclosure	ob-labs/contextseek
langchain-oceanbase	Data substrate — checkpoint + store + vector + hybrid search on OceanBase / seekdb / MySQL	oceanbase/langchain-oceanbase

Quick Start — for LangChain developers¶

Which template should I pick?

Starting fresh / learning? → langchain/markdown-messages (minimal, 5 min)
Already have a graph, need to deliver a product? → langchain/default (frontend + Feishu IM + full runtime)
Need deep research with sub-agents? → deepagents/research (Tavily + report generation)
Graph runs on a remote server (agentseek-api / LangSmith)? → langchain/cli-remote

# Pick one and run:
uvx --from agentseek-cli agentseek create langchain --template markdown-messages
# or: langchain --template default
# or: deepagents --template research

Then: cd <project> && uv sync && uv run langgraph dev (minimal) or uv run agentseek run (full delivery).

LangSmith tracing is pre-configured. Every template ships a .env.example with LANGSMITH_TRACING=true and LANGSMITH_API_KEY ready to fill in. Set your key and you get full run observability in LangSmith immediately.

Next steps after your agent runs:

Add persistent memory → ContextSeek docs
Ship to production as a service → agentseek-api docs
Switch to a durable database → langchain-oceanbase docs
Install dev skills for guided help → see Development skills below
Study DeepAgents systematically → see Open-source course below
Understand the full relationship → How agentseek relates to LangChain

For OceanBase / seekdb / MySQL developers¶

Already running OceanBase, seekdb, or MySQL? AgentSeek uses your database as the data substrate for AI agents — checkpoint, persistent memory, vector search, and hybrid retrieval all run on the DB you already operate.

pip install langchain-oceanbase[pyseekdb]   # OceanBase / seekdb
pip install langchain-oceanbase             # MySQL (checkpoint + store)

This gives you:

LangGraph Checkpoint — durable execution state for long-running agents
Store — cross-session persistent memory (namespaced key-value)
VectorStore + Hybrid Search — embedding retrieval fused with BM25 (OceanBase / seekdb)

MySQL users get checkpoint and store out of the box; vector search requires OceanBase or seekdb. Either way, runtime data is queryable SQL from day one.

Get started:

Install: pip install langchain-oceanbase[pyseekdb]
Read the integration guide: langchain-oceanbase README
Pick a template to run a full agent on top: agentseek create langchain --template default
For the harness tape store plugin: see agentseek-tapestore-oceanbase

New to agents? Start here¶

Never built an AI agent before? No problem.

Make sure you have Python 3.12+ and uv installed
Get a model API key (OpenRouter free tier works: openrouter.ai)
Run:

uvx --from agentseek-cli agentseek create langchain --template markdown-messages
cd markdown_messages_agent
cp .env.example .env   # fill in your API key
uv sync && uv run langgraph dev

You now have a chatbot running locally. Open the URL printed in the terminal.

Where to go from here:

Quick demo tutorial — 5-minute walkthrough with screenshots
Build your first harness app — full tutorial with frontend

The suite in action¶

agentseek-api — ship your graph to production¶

uv run agentseek-api dev
curl http://127.0.0.1:2024/info

Implements the Agent Protocol (threads, runs, streaming, Store API, MCP, A2A). Your LangGraph code runs unchanged behind standard HTTP endpoints.

Full docs: github.com/ob-labs/agentseek-api

ContextSeek — semantic context layer¶

from contextseek import ContextSeek

ctx = ContextSeek.from_settings()
ctx.add("OceanBase is a distributed database for financial workloads",
        scope="acme/db", source="wiki")

for hit in ctx.retrieve("distributed database", scope="acme/db", k=5):
    print(hit.item.stage, hit.score, hit.item.summary[:60])

Unified ContextItem model with provenance, L0/L1/L2 progressive disclosure, EvolutionEngine, DreamEngine. Accessible via HTTP, MCP, or Python SDK. Ships with a LangChain middleware for automatic context injection per turn and LangSmith @traceable support for full observability.

Full docs: github.com/ob-labs/contextseek

langchain-oceanbase — the data substrate¶

pip install langchain-oceanbase[pyseekdb]

LangGraph Checkpoint + Store + VectorStore + Hybrid Search — all on one database (OceanBase, seekdb, or MySQL). Runtime data is queryable SQL from day one.

Full docs: github.com/oceanbase/langchain-oceanbase

Open-source course¶

"Deep Agents 实战" — a free course on building production-grade AI agents with LangChain / DeepAgents. All hands-on labs use AgentSeek.

Course site · Source repo

Topics covered: Agent Harness concepts, virtual filesystem, task planning, sub-agents, async delegation, long-term memory, Human-in-the-Loop, skills, sandboxes, streaming frontends, and production deployment.

Development skills¶

AgentSeek ships a set of development skills — installable guides that live inside your AI coding agent (Claude Code, Cursor, etc.) and help you build LangChain applications without leaving the editor.

Skill	What it does
langchain-dev-guide	Engineering pitfalls and verified fixes for LangChain / LangGraph. Covers DeepAgents, middleware, streaming, multi-agent orchestration, and common issues — each with Symptom → Cause → Solution.
langchain-cn-models	Step-by-step recipes for integrating Chinese LLM providers (DeepSeek, Qwen, GLM, Moonshot, etc.) into LangChain via the OpenAI-compatible interface.

Install all skills at once:

npx skills add ob-labs/agentseek --all

Or pick specific ones:

npx skills add ob-labs/agentseek --skill langchain-dev-guide --agent claude-code
npx skills add ob-labs/agentseek --skill langchain-cn-models --agent claude-code

Once installed, your coding agent can reference these guides when you hit a LangChain issue — no manual doc searching needed.

Full details: skills/ | How to add skills: Add skills guide

Connect your Agent Framework¶

AgentSeek is designed to be the harness underneath any Agent Framework — not just LangChain. If you are building a new Agent Framework or maintaining one that needs a durable data layer and semantic context, we welcome you to connect it. Bub is a good example — it ships built-in as AgentSeek's native framework through exactly this integration pattern.

What AgentSeek brings to your framework:

Data substrate — checkpoint, persistent memory, vector search, and hybrid retrieval on OceanBase / seekdb / MySQL. Your agents get durable, queryable runtime data from day one without you building a storage layer.
Semantic context layer — ContextSeek handles memory accumulation, retrieval, progressive disclosure, and evolution. Your framework gets cross-session intelligence for free.
Production serving — agentseek-api implements Agent Protocol. Your framework's runnables can serve behind standard HTTP endpoints.
IM delivery & templates — Feishu / DingTalk / Slack gateways and cookiecutter project scaffolding, ready for your framework to plug into.

How to integrate:

The integration pattern is the same one agentseek-langchain follows — a contrib plugin that bridges your framework's runnable into the harness turn pipeline. See The extension model and How to author a contrib plugin.

We'd love to collaborate — open an issue or a PR under contrib/.

Other paths¶

Already using Bub? AgentSeek is a distribution of Bub with opinionated defaults. Try agentseek create bub --template default for CopilotKit + Feishu without LangChain. See How agentseek relates to Bub.

Want the raw harness CLI? See Choosing an entry point.