AgentHub
A self-hosted, multi-agent platform where an orchestrator decomposes a goal into roles, dispatching implementation, code-review and verification agents that work together and stream their progress.
AgentHub is a self-hosted platform for running teams of AI agents that coordinate on a task rather than acting as a single chat-based agent. An orchestrator agent decomposes a goal and dispatches roles to other agents; code-review and verification agents check the work, and deployment and implementation agents finish it, with every agent running under a bounded, observable loop that streams its actions and logs.
Running the full platform locally starts PostgreSQL through Docker along with the backend and frontend, and a built-in mock provider lets it run offline without an API key; OpenAI-compatible, Anthropic or other model keys can be set as environment variables to use a real provider, optionally routed through a unified gateway. A separate CLI can be installed into an isolated Python virtual environment for local use, with Windows x64 binaries published as prebuilt releases.
The stack is built from Go, Python and Rust microservices with a Next.js admin UI, and is Apache-2.0 licensed.