OpenFang
www.openfang.sh- Category
- AI Agents
- Rank
- No. 240Tools index
- Pricing
- Open Source
- Platform
- desktop · cli
- Type
- APP
- Builder
- @RightNow-AI
- GitHub
- 18.1k stars
- Latest release
- v0.6.9
- Added
- Mar 6, 2026
About
An open-source Agent Operating System built in Rust that provides autonomous AI agents with advanced security, multi-platform channel support, and pre-built capabilities. Features 7 autonomous 'Hands' that can perform tasks like video editing, lead generation, and web automation with 16 security layers and 40 communication channels.
What it does
OpenFang runs persistent agents on schedules, routes their work through configurable tools and models, and exposes results through a dashboard, messaging channels, APIs, and desktop software. Agent packages combine a manifest, operational instructions, domain guidance, and approval rules. A workflow engine chains agents sequentially, in parallel, conditionally, or through bounded loops.
Why it's ranked here
This is a serious systems project, not merely a prompt collection. The Rust workspace covers runtime, memory, APIs, channels, desktop delivery, skills, and extensions. Security controls and protocol interoperability receive unusually concrete documentation. The main reservation is maturity: the project explicitly warns that its pre-1.0 releases may have rough edges and breaking changes.
What's good
Deployment centers on one compiled binary, while the same kernel supports a browser dashboard, command line operation, and a native desktop wrapper. Agents can consume external MCP tools or appear as MCP tools themselves. Capability checks, metered WebAssembly execution, tamper-evident auditing, secret zeroization, SSRF defenses, and purchase approval gates address distinct failure modes rather than relying on one sandbox.
Tradeoffs
The project calls itself feature complete but still pre-1.0 and recommends pinning a commit for production. Its documentation also contains inconsistent inventory figures for tests, tools, providers, and crate structure, so published counts require verification. Some capabilities need external programs or services, including browser automation, media tooling, messaging gateways, and model providers. The WhatsApp bridge adds a separate JavaScript process and persistent authentication state.
How to use it well
Use OpenFang when you want long-running, self-hosted agent jobs with schedules, approvals, messaging delivery, and observable multi-step workflows. Start with one bounded task, grant narrow capabilities, select a model provider, and inspect outputs before adding automation. It suits operators comfortable managing credentials and local services. It does not replace a lightweight application library or a simple conversational chatbot framework.
Technical notes+
Cargo.toml defines a Rust 2021 workspace with 14 members, Rust 1.75, Axum, Tokio, SQLite through rusqlite, Wasmtime, Tauri, and the rmcp SDK; its workspace license is Apache-2.0 OR MIT. docs/mcp-a2a.md describes bidirectional MCP support, with stdio and SSE clients, a stdio server from openfang mcp, and an HTTP endpoint at POST /mcp. docs/security.md maps capability enforcement, dual WASM metering, Merkle-linked auditing, taint tracking, SSRF controls, and other defenses to specific crates. docs/desktop.md documents an embedded Axum server inside a Tauri 2.0 shell. packages/whatsapp-gateway/index.js implements a separate Baileys-based HTTP bridge that forwards messages to the agent API and stores multi-file authentication state.
Observed
- License
- The Rust workspace declares Apache-2.0 OR MIT.
- Primary language
- Rust 2021 workspace, with a separate JavaScript WhatsApp gateway.
- Packaging
- Single compiled binary plus native Tauri desktop installers.
- Interfaces
- CLI, browser dashboard, REST, WebSocket, SSE, MCP client and server, and A2A integration.
- Platform support
- Desktop builds target Windows, macOS, and Linux.
- Storage
- SQLite-backed memory with bundled SQLite support.
- Workspace structure
- Cargo workspace lists 14 members covering types, memory, runtime, wire, API, kernel, CLI, channels, migration, skills, desktop, hands, extensions, and build tasks.
Read from README.md, Cargo.toml, packages/whatsapp-gateway/index.js, docs/README.md, docs/desktop.md, docs/mcp-a2a.md, docs/security.md, docs/providers.md, docs/workflows.md.
What it can do
Edit videos autonomously
Video files and editing instructions → Edited video content
Generate leads automatically
Target criteria and search parameters → List of potential leads with contact information
Automate web tasks
Web automation scripts and target websites → Completed web interactions and extracted data
Deploy autonomous AI agents
Agent configuration and task definitions → Running autonomous agents performing specified tasks
Secure agent communications
Agent messages and data → Encrypted and secured communications across 16 security layers
Manage multi-platform channels
Channel configurations and connection parameters → Active connections across 40 communication channels
Build custom agent applications
Rust code and agent specifications → Compiled agent applications with 14 available crates
Tags
Tech Stack
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.