What it does

OpenFang is an open-source agent operating system written in Rust and compiled into a single binary of about 32 MB. Rather than waiting for a prompt, its agents can run on schedules and report results to a dashboard or a messaging channel. After installing, you run init to set up a provider, start the daemon, and open the dashboard on localhost. Its main building block is the Hand, a packaged autonomous capability with a manifest, a system prompt, a skill reference and approval guardrails. Seven are bundled: Clip, Lead, Collector, Predictor, Researcher, Twitter and Browser. The Browser Hand requires approval before any purchase, and the Twitter Hand uses an approval queue before posting. You can write your own Hand and publish it to the FangHub skill marketplace. The page lists 40 channel adapters, 27 LLM providers through three native drivers, an OpenAI-compatible API, MCP and A2A support, and 16 security systems including a WASM sandbox and a hash-chained audit trail. A desktop app and a migration command for OpenClaw users are also listed. The README states the project is pre-1.0 (v0.5.10), expects breaking changes between minor versions, and advises pinning a commit for production.

Capabilities

Single Rust binary of about 32 MB with a local dashboard
Seven bundled Hands that run on schedules without prompting
40 messaging channel adapters
27 LLM providers with fallback routing and cost tracking
OpenAI-compatible API plus MCP and A2A support
WASM sandbox, audit trail and approval gates

When to use it

You want agents that work on a schedule, such as research or lead generation, without prompting each run
You want to self-host agents and reach them from many messaging apps
You are moving from OpenClaw and want a migration command
You accept pre-1.0 software and will pin versions