LLMling-Agent
AgentPool is a unified agent orchestration hub that lets you configure and manage heterogeneous AI agents via YAML and expose them through standardized protocols. It solves the problem of coordinating multiple AI agents with different APIs and protocols by acting as a protocol bridge. It is for developers who want to integrate and manage various AI agents (like Claude Code, Codex, Goose) in a unified way.
✨ Key features
- YAML-based configuration for all agents
- Supports native, direct, ACP, and AG-UI agents
- Inter-agent delegation and message routing
- Team and chain orchestration modes
- Multiple server protocols: ACP, OpenCode, MCP, AG-UI, OpenAI API
- Structured output, storage, triggers, and streaming TTS
🎯 Use cases
- Coordinate multiple AI agents for complex coding tasks
- Expose agents to IDEs via ACP server
- Build sequential or parallel agent pipelines
- Integrate remote agents via OpenCode server
- Use as a drop-in OpenAI API replacement
📦 Installation
🧰 Requirements: Requires Python and uv tool installation; no API keys mentioned in README.
uv tool install agentpool
🚀 Usage
# agents.yml
agents:
assistant:
type: native
model: openai:gpt-4o
system_prompt: "You are a helpful assistant."
# Run via CLI
agentpool run assistant "Hello!"
⚠️ Good to know
No limitations or project status mentioned in the README.
❓ FAQ
What agent types can I configure?
You can configure native PydanticAI-based agents, direct integrations like Claude Code and Codex, external ACP agents like Goose, and AG-UI agents.
How do I expose agents to IDEs?
Use the ACP server command agentpool serve-acp agents.yml to expose agents to ACP clients like Zed and Toad.
Can agents work together?
Yes, agents can form teams (parallel) or chains (sequential) and delegate work to each other via the subagent tool.
What server protocols are supported?
AgentPool supports ACP, OpenCode, MCP, AG-UI, and OpenAI API server protocols.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.