AgentSeek

中文 | English

AgentSeek is a template-first toolkit for local agent application development. It gives editable generated projects one predictable lifecycle: discover, create, inspect, configure, check, run, observe, and iterate.

AgentSeek 0.1.1 resolves lifecycle-v2 templates from the immutable agentseek-ai/agentseek-templates catalog. The CLI lists templates from its embedded registry snapshot and fetches named template content at the exact locked commit.

Experience the local ADLC

Start with the shipped deepagents/research walkthrough. It creates a fully editable DeepAgents research app with a native LangGraph backend and a React frontend as its primary user experience.

# Install the lifecycle CLI, then create and enter the generated project.
uv tool install agentseek
agentseek create deepagents/research --no-input
cd research_deepagent

# Inspect declared entry points before configuring local credentials.
agentseek info
cp .env.example .env
cp frontend/.env.example frontend/.env
$EDITOR .env

# Run the setup tasks declared by this project, then check readiness.
agentseek task --list
agentseek task sync
agentseek task frontend
agentseek doctor

# Preview the native commands, run the local stack, and check live services.
agentseek dev --dry-run
agentseek dev
# In another terminal, after agentseek dev starts, check live services.
agentseek doctor --live

The current scaffold declares only sync and frontend setup tasks. Its native backend resolves to uv run langgraph dev --port 2024 --no-browser, while the frontend resolves to npm run dev in frontend/.

For a one-off run, use uvx agentseek create deepagents/research --no-input instead of installing the CLI.

What is AgentSeek?

AgentSeek supplies the stable lifecycle surface around a generated project; it does not own that project's source, framework, runtime, or deployment. A template selects those pieces, and the generated project remains yours to edit.

The CLI connects people, coding agents, and desktop clients to a locked, versioned template catalog and an editable project lifecycle contract. The project then owns its runtimes and integrations, including models, tools, MCP servers, and external services.

Agent Development Lifecycle

The local ADLC keeps iteration anchored in the existing project instead of starting over: discover, create, inspect, configure, check, run, observe, and iterate back to inspect.

Observability throughout the loop

Lifecycle diagnostics answer whether the declared project is ready and running: agentseek info shows topology, agentseek doctor performs preflight checks, and agentseek doctor --live checks declared services after startup. For Desktop, scripts, and other machine consumers, use the versioned JSON contract instead of parsing human output.

agentseek info --json
agentseek doctor --json
agentseek doctor --live --json
AGENTSEEK_CONSOLE=true agentseek doctor --live

info --json contains normalized services, references, and safe actions such as which URL can be opened directly. It never includes environment values or raw process or task commands. AGENTSEEK_CONSOLE=true enables local CLI spans and lifecycle events. Optional LangSmith tracing, configured through the settings already present in deepagents/research, answers what the agent did inside an individual run.

Guided templates

Choose by goal, then inspect a specific template before creating it. AgentSeek maintains many templates without requiring you to memorize a catalog dump.

agentseek create --list-templates
agentseek create --list-templates --filter deepagents
agentseek create deepagents/research --describe

Core concepts and commands

Documentation

• Documentation home
• Get started
• Guides
• Reference
• Concepts

Development

git clone https://github.com/ob-labs/agentseek.git
cd agentseek
make install
make check
make test
make docs-test

Community and course

Explore Deep Agents in Action, a free LangChain / DeepAgents course with AgentSeek labs, in the course repository. Join or follow project discussion through GitHub Discussions.

License

Apache-2.0