AgentPool is a unified AI agent orchestration hub used to configure and manage multiple heterogeneous AI agents and expose them externally through standardized protocols.
[Main problem solved] If users want to use multiple AI agents at the same time—for example, using Claude Code for refactoring, using Codex for code editing with advanced reasoning, using a custom analysis agent, or using Goose for specific tasks—they face the problem that each agent has different APIs, protocols, and integration methods, and coordinating them requires writing glue code for each combination. AgentPool acts as a protocol bridge, defining all agents in a single YAML file and then exposing them through a unified interface.
[Supported agent types]
- Native agents (based on PydanticAI)
- Direct integrations (Claude Code, Codex)
- External ACP agents (such as Goose)
- AG-UI agents
[Main capabilities]
- Centrally define all agents in a single YAML file, minimizing configuration
- Mix multiple agent types; a coordinator can delegate work to any agent, and all are accessed through the same interface
- Multi-agent collaboration: agents can form teams (parallel) or chains (sequential)
- Rich YAML configuration options: models, tools, connections, triggers, and storage are all configurable
- Server protocols: agents can be exposed through multiple server protocols. The ACP server is suitable for IDE integration, providing real-time tool confirmation and session management; the OpenCode server enables the OpenCode TUI to control AgentPool agents, including agents running in remote environments (Docker, SSH, cloud sandboxes)
- Structured output: response schemas can be defined inline or Python types can be imported
- Storage and analytics: track all interactions through configurable providers
- File abstraction: UPath-based operations support local and remote sources
- Triggers: can respond to file changes, webhooks, or custom events
- Streaming TTS: provides voice output support for all agents
[Usage] Define agents through a YAML file (including native, direct integration, external ACP, AG-UI, and other types), then expose them externally through the ACP or AG-UI protocol. Specific installation commands, startup commands, and code examples are not currently provided in the supplied materials and require reference to the full documentation. Full documentation (including advanced configuration, connection modes, and API reference) is available at phil65.github.io/agentpool.
[Applicable scenarios]
- Need to coordinate multiple different AI agents at the same time to complete complex tasks
- Use agents through IDE integration (ACP server)
- Control agents through the OpenCode TUI, including agents in remote environments
- Workflows that require multi-agent parallel collaboration or sequential chained execution
- Scenarios that require structured output, interaction tracking, file operations, event triggers, or voice output
[Limitations] The materials do not provide specific license agreements, pricing information, platform requirements, or performance data.
