RoboFang is an orchestration layer for local AI agents, designed to bridge a federated fleet of MCP servers with physical robotics and virtual environments.
Standard technical documentation is organized by category:
| Domain | Description |
|---|---|
| Architecture | Multi-process topology and module layout. |
| Setup | Local deployment guide (UV, Ollama, Dashboard). |
| Council Protocol | Multi-model coordination and adjudication logic. |
| Fleet Management | Port registry and connector integration for the 30+ server fleet. |
| Autonomous Hands | Manifest-driven background processes. |
| Comparison Matrix | Technical mapping against OpenClaw and OpenFang. |
| Roadmap | Strategic development phases and priorities. |
| Status | Operational health and tech debt registry. |
- Backend: FastAPI + FastMCP 3.1+ (Python 3.12).
- Frontend: React + Vite (Port 10864).
- Adjudication: Council of Dozens (Foreman/Worker/Satisficer logic).
- Inference: Local Ollama (Primary).
- Local-First: Zero-token cost operations via optimized local models.
- Unified Tool Surface: Standardized API gateway for heterogeneous MCP backends.
- Embodiment: Joint control for robotics (Bumi, Yahboom) and VR systems (Resonite, VRChat).
- Security Gates: Policy-driven adversarial adjudication for sensitive operations.
- Environment:
uv sync - Backend:
uv run python -m robofang.cli start bridge - Frontend:
cd dashboard; .\start.ps1
Visit http://localhost:10864 to manage the coordination node.