OpenGoat builds hierarchical "organizations" of AI agents on top of OpenClaw, coordinating work across multiple coding tools including Claude Code, Codex, Cursor, GitHub Copilot CLI, and Lovable. Instead of a single assistant, you create a reporting structure: a default entry agent ("Goat", framed as an AI co-founder) plus manager and individual-contributor agents created with roles like CTO, engineer, or designer, each declaring who they report to and which skills they carry. Work is then addressed to roles — asking the CTO agent to plan a roadmap, or an engineer agent to implement a sprint item — with named sessions preserving continuity across messages.
It ships as a CLI and a local web UI. Projects get their own scoped agent workspaces (creating a project provisions a CMO agent bootstrapped from the project URL), and a task system lets agents create, assign, and track tasks between roles. Skills are installable per agent, with role-specific board skills for managers versus individual contributors on OpenClaw agents. It can run against a local OpenClaw install or an external OpenClaw gateway, and also via Docker.
