OpenFang is a comprehensive agent operating system designed for building, running, and deploying autonomous AI agents. Built entirely in Rust with 14 crates and 137K lines of code, it offers kernel-grade architecture with zero clippy warnings. The system features 7 pre-built autonomous Hands that work on schedules, 30 pre-built agents across 4 performance tiers, 38 built-in tools plus Model Context Protocol (MCP) support, and 40 channel adapters for platforms like Telegram, Discord, Slack, and WhatsApp.
Key Features:
- Autonomous Hands: 7 capability packages (Clip, Lead, Collector, Predictor, Researcher, Twitter, Browser) that run on schedules and report to dashboards
- Security Systems: 16 layers including WASM dual-metered sandbox, Merkle audit trail, taint tracking, SSRF protection, and HMAC-SHA256 mutual authentication
- Runtime: Sandboxed execution with WASM, workspace-confined file operations, and 10-phase graceful shutdown
- Memory: SQLite-backed storage with vector embeddings and cross-channel canonical sessions
- Protocols: MCP (client + server), Google A2A agent-to-agent tasks, and OpenFang Protocol for P2P networking
- Desktop: Tauri 2.0 native application with system tray, notifications, and global shortcuts
Use Cases:
- Content creation and social media management (Clip, Twitter Hands)
- Lead generation and business intelligence (Lead, Collector Hands)
- Research and forecasting (Researcher, Predictor Hands)
- Web automation and workflow completion (Browser Hand)
- Multi-platform chatbot deployment across 40+ channels
- Secure agent development with comprehensive audit trails
Technical Highlights:
- Single binary deployment for macOS, Linux, and Windows
- 53 tools including web search, browser automation, image generation, TTS, Docker, and knowledge graphs
- 27 LLM providers including Anthropic, Gemini, Groq, and DeepSeek
- Benchmarks show 180ms cold start time, 40MB idle memory usage, and 32MB install size
- MIT licensed with full open-source availability
