Home

Connect all the agents!

Key Features¶

Slash Commands¶

Skills exposed as slash commands across all supported protocols (ACP, AG-UI, OpenCode):

Define reusable skill instructions in SKILL.md files
Automatically exposed as protocol-native commands
Use /skill:my-skill in OpenCode, skill__my-skill tool in AG-UI, or slash commands in ACP

ACP Integration¶

First-class support for the Agent Client Protocol (ACP):

Integrate directly into IDEs like Zed, VS Code, and others
Wrap external agents (Claude Code, Goose, Codex, fast-agent) as nodes
Unified node abstraction - ACP agents work like native agents
Compose ACP agents into teams with native agents

📝 Easy Agent Configuration¶

AgentPool excels at static YAML-based agent configuration:

Define agents with extreme detail in pure YAML (Pydantic-backed)
Expansive JSON schema for IDE autocompletion and validation, backed by an extremely detailed schema.
Multi-Agent setups with native as well as remote (ACP / AGUI) agents

🧩 Unified Node Architecture¶

Everything is a MessageNode - enabling seamless composition:

Native agents with a large set of default tools
ACP agents
AG-UI agents
Teams (parallel and sequential)
Human-in-the-loop-agents
All nodes share the same interface

Dependencies¶

Category	Representative Packages
Framework & AI	`pydantic`, `pydantic-ai-slim`, `pydantic-graph`
Web, Server & Protocols	`fastapi`, `mcp`, `starlette`, `uvicorn`, `websockets`
Storage & Database	`sqlalchemy`, `sqlmodel`, `alembic`
CLI & Configuration	`typer`, `rich`, `yamling`, `schemez`
Async, IO & Execution	`anyio`, `anyenv`, `fsspec`, `watchfiles`
Observability	`logfire`, `structlog`
Documents & Search	`docler`, `searchly`, `ripgrep-rs`, `tokonomics`
Tooling & Events	`jinja2`, `psygnal`, `evented`, `slashed`, `pydocket`

See the full dependency list for complete details and version info.

License¶

MIT License - see LICENSE for details.

Documentation¶

Tutorials — Getting started and learning guides
How-To Guides — Task-oriented guides for configuration, servers, and advanced features
Reference — CLI commands, core concepts, and API reference
Architecture — How and why AgentPool works
Decision Records — Architecture Decision Records (ADRs)
RFCs — Request for Comments proposals and status
Documentation Guide — Where to put new documentation

Quick Start¶

Basic Agent Configuration¶

# agents.yml
agents:
  assistant:
    display_name: "Technical Assistant"
    model: openai:gpt-4
    system_prompt: You are a helpful technical assistant.
    tools:
      - type: file_access

Python Usage¶

from agentpool import AgentPool

async def main():
    async with AgentPool("agents.yml") as pool:
        agent = pool.get_agent("assistant")
        response = await agent.run("What is Python?")
        print(response.data)

if __name__ == "__main__":
    import anyio
    anyio.run(main)