OpenFang
The Agent Operating System
Open-source Agent OS built in Rust. 137K LOC. 14 crates. 1,767+ tests. Zero clippy warnings.
One binary. Battle-tested. Agents that actually work for you.
Documentation •
Quick Start •
Twitter / X
v0.5.10 (April 2026)
OpenFang is feature complete but still pre-1.0. Expect rough edges and breaking changes between minor versions. We ship fast and fix fast. Pin to a specific commit for production use until v1.0. Report issues here.
What is OpenFang?
OpenFang is an open-source Agent Operating System. Not a chatbot framework. Not a Python wrapper around an LLM. Not a "multi-agent orchestrator." A full operating system for autonomous agents, built from scratch in Rust.
Traditional agent frameworks wait for you to type something. OpenFang runs autonomous agents that work for you: on schedules, 24/7, building knowledge graphs, monitoring targets, generating leads, managing your social media, and reporting results to your dashboard.
The entire system compiles to a single ~32MB binary. One install, one command, your agents are live.
curl -fsSL https://openfang.sh/install | sh
openfang init
openfang start
# Dashboard live at http://localhost:4200
Windows
irm https://openfang.sh/install.ps1 | iex
openfang init
openfang start
Hands: Agents That Actually Do Things
"Traditional agents wait for you to type. Hands work for you."
Hands are OpenFang's core innovation. Pre-built autonomous capability packages that run independently, on schedules, without you having to prompt them. This is not a chatbot. This is an agent that wakes up at 6 AM, researches your competitors, builds a knowledge graph, scores the findings, and delivers a report to your Telegram before you've had coffee.
Each Hand bundles:
- HAND.toml: manifest declaring tools, settings, requirements, and dashboard metrics.
- System Prompt: multi-phase operational playbook. Not a one-liner. These are 500+ word expert procedures.
- SKILL.md: domain expertise reference injected into context at runtime.
- Guardrails: approval gates for sensitive actions (e.g. Browser Hand requires approval before any purchase).
All compiled into the binary. No downloading, no pip install, no Docker pull.
The 7 Bundled Hands
| Hand |
What It Actually Does |
| Clip |
Takes a YouTube URL, downloads it, identifies the best moments, cuts them into vertical shorts with captions and thumbnails, optionally adds AI voice-over, and publishes to Telegram and WhatsApp. 8-phase pipeline. FFmpeg + yt-dlp + 5 STT backends. |
| Lead |
Runs daily. Discovers prospects matching your ICP, enriches them with web research, scores 0-100, deduplicates against your existing database, and delivers qualified leads in CSV/JSON/Markdown. Builds ICP profiles over time. |
| Collector |
OSINT grade intelligence. You give it a target (company, person, topic). It monitors continuously: change detection, sentiment tracking, knowledge graph construction, and critical alerts when something important shifts. |
| Predictor |
Superforecasting engine. Collects signals from multiple sources, builds calibrated reasoning chains, makes predictions with confidence intervals, and tracks its own accuracy using Brier scores. Has a contrarian mode that deliberately argues against consensus. |
| Researcher |
Deep autonomous researcher. Cross-references multiple sources, evaluates credibility using CRAAP criteria (Currency, Relevance, Authority, Accuracy, Purpose), generates cited reports with APA formatting, supports multiple languages. |
| Twitter |
Autonomous Twitter/X account manager. Creates content in 7 rotating formats, schedules posts for optimal engagement, responds to mentions, tracks performance metrics. Has an approval queue, so nothing posts without your OK. |
| Browser |
Web automation agent. Navigates sites, fills forms, clicks buttons, handles multi-step workflows. Uses Playwright bridge with session persistence. Mandatory purchase approval gate: it will never spend your money without explicit confirmation. |
# Activate the Researcher Hand. It starts working immediately.
openfang hand activate researcher
# Check its progress anytime
openfang hand status researcher
# Activate lead generation on a daily schedule
openfang hand activate lead
# Pause without losing state
openfang hand pause lead
# See all available Hands
openfang hand list
Build your own. Define a HAND.toml with tools, settings, and a system prompt. Publish to FangHub.
OpenFang vs The Landscape
Benchmarks: Measured, Not Marketed
All data from official documentation and public repositories, February 2026.
Cold Start Time (lower is better)
ZeroClaw ██░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 10 ms
OpenFang ██████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 180 ms ★
LangGraph █████████████████░░░░░░░░░░░░░░░░░░░░░░░░░ 2.5 sec
CrewAI ████████████████████░░░░░░░░░░░░░░░░░░░░░░ 3.0 sec
AutoGen ██████████████████████████░░░░░░░░░░░░░░░░░ 4.0 sec
OpenClaw █████████████████████████████████████████░░ 5.98 sec
Idle Memory Usage (lower is better)
ZeroClaw █░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 5 MB
OpenFang ████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 40 MB ★
LangGraph ██████████████████░░░░░░░░░░░░░░░░░░░░░░░░░ 180 MB
CrewAI ████████████████████░░░░░░░░░░░░░░░░░░░░░░░ 200 MB
AutoGen █████████████████████████░░░░░░░░░░░░░░░░░░ 250 MB
OpenClaw ████████████████████████████████████████░░░░ 394 MB
Install Size (lower is better)
ZeroClaw █░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 8.8 MB
OpenFang ███░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 32 MB ★
CrewAI ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 100 MB
LangGraph ████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 150 MB
AutoGen ████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░ 200 MB
OpenClaw ████████████████████████████████████████░░░░ 500 MB
Security Systems (higher is better)
OpenFang ████████████████████████████████████████████ 16 ★
ZeroClaw ███████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 6
OpenClaw ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 3
AutoGen █████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 2
LangGraph █████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 2
CrewAI ███░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 1
Channel Adapters (higher is better)
OpenFang ████████████████████████████████████████████ 40 ★
ZeroClaw ███████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 15
OpenClaw █████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 13
CrewAI ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 0
AutoGen ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 0
LangGraph ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 0
LLM Providers (higher is better)
ZeroClaw ████████████████████████████████████████████ 28
OpenFang ██████████████████████████████████████████░░ 27 ★
LangGraph ██████████████████████░░░░░░░░░░░░░░░░░░░░░ 15
CrewAI ██████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 10
OpenClaw ██████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 10
AutoGen ███████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 8
Feature-by-Feature Comparison
| Feature |
OpenFang |
OpenClaw |
ZeroClaw |
CrewAI |
AutoGen |
LangGraph |
| Language |
Rust |
TypeScript |
Rust |
Python |
Python |
Python |
| Autonomous Hands |
7 built-in |
None |
None |
None |
None |
None |
| Security Layers |
16 discrete |
3 basic |
6 layers |
1 basic |
Docker |
AES enc. |
| Agent Sandbox |
WASM dual-metered |
None |
Allowlists |
None |
Docker |
None |
| Channel Adapters |
40 |
13 |
15 |
0 |
0 |
0 |
| Built-in Tools |
53 + MCP + A2A |
50+ |
12 |
Plugins |
MCP |
LC tools |
| Memory |
SQLite + vector |
File-based |
SQLite FTS5 |
4-layer |
External |
Checkpoints |
| Desktop App |
Tauri 2.0 |
None |
None |
None |
Studio |
None |
| Audit Trail |
Merkle hash-chain |
Logs |
Logs |
Tracing |
Logs |
Checkpoints |
| Cold Start |
<200ms |
~6s |
~10ms |
~3s |
~4s |
~2.5s |
| Install Size |
~32 MB |
~500 MB |
~8.8 MB |
~100 MB |
~200 MB |
~150 MB |
| License |
MIT |
MIT |
MIT |
MIT |
Apache 2.0 |
MIT |
16 Security Systems: Defense in Depth
OpenFang doesn't bolt security on after the fact. Every layer is independently testable and operates without a single point of failure.
| # |
System |
What It Does |
| 1 |
WASM Dual-Metered Sandbox |
Tool code runs in WebAssembly with fuel metering + epoch interruption. A watchdog thread kills runaway code. |
| 2 |
Merkle Hash-Chain Audit Trail |
Every action is cryptographically linked to the previous one. Tamper with one entry and the entire chain breaks. |
| 3 |
Information Flow Taint Tracking |
Labels propagate through execution. Secrets are tracked from source to sink. |
| 4 |
Ed25519 Signed Agent Manifests |
Every agent identity and capability set is cryptographically signed. |
| 5 |
SSRF Protection |
Blocks private IPs, cloud metadata endpoints, and DNS rebinding attacks. |
| 6 |
Secret Zeroization |
Zeroizing<String> auto-wipes API keys from memory the instant they're no longer needed. |
| 7 |
OFP Mutual Authentication |
HMAC-SHA256 nonce-based, constant-time verification for P2P networking. |
| 8 |
Capability Gates |
Role based access control. Agents declare required tools, the kernel enforces it. |
| 9 |
Security Headers |
CSP, X-Frame-Options, HSTS, X-Content-Type-Options on every response. |
| 10 |
Health Endpoint Redaction |
Public health check returns minimal info. Full diagnostics require authentication. |
| 11 |
Subprocess Sandbox |
env_clear() + selective variable passthrough. Process tree isolation with cross-platform kill. |
| 12 |
Prompt Injection Scanner |
Detects override attempts, data exfiltration patterns, and shell reference injection in skills. |
| 13 |
Loop Guard |
SHA256-based tool call loop detection with circuit breaker. Handles ping-pong patterns. |
| 14 |
Session Repair |
7-phase message history validation and automatic recovery from corruption. |
| 15 |
Path Traversal Prevention |
Canonicalization with symlink escape prevention. ../ doesn't work here. |
| 16 |
GCRA Rate Limiter |
Cost-aware token bucket rate limiting with per-IP tracking and stale cleanup. |
Architecture
14 Rust crates. 137,728 lines of code. Modular kernel design.
openfang-kernel Orchestration, workflows, metering, RBAC, scheduler, budget tracking
openfang-runtime Agent loop, 3 LLM drivers, 53 tools, WASM sandbox, MCP, A2A
openfang-api 140+ REST/WS/SSE endpoints, OpenAI-compatible API, dashboard
openfang-channels 40 messaging adapters with rate limiting, DM/group policies
openfang-memory SQLite persistence, vector embeddings, canonical sessions, compaction
openfang-types Core types, taint tracking, Ed25519 manifest signing, model catalog
openfang-skills 60 bundled skills, SKILL.md parser, FangHub marketplace
openfang-hands 7 autonomous Hands, HAND.toml parser, lifecycle management
openfang-extensions 25 MCP templates, AES-256-GCM credential vault, OAuth2 PKCE
openfang-wire OFP P2P protocol with HMAC-SHA256 mutual authentication
openfang-cli CLI with daemon management, TUI dashboard, MCP server mode
openfang-desktop Tauri 2.0 native app (system tray, notifications, global shortcuts)
openfang-migrate OpenClaw, LangChain, AutoGPT migration engine
xtask Build automation
40 Channel Adapters
Connect your agents to every platform your users are on.
Core: Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Email (IMAP/SMTP)
Enterprise: Microsoft Teams, Mattermost, Google Chat, Webex, Feishu/Lark, Zulip
Social: LINE, Viber, Facebook Messenger, Mastodon, Bluesky, Reddit, LinkedIn, Twitch
Community: IRC, XMPP, Guilded, Revolt, Keybase, Discourse, Gitter
Privacy: Threema, Nostr, Mumble, Nextcloud Talk, Rocket.Chat, Ntfy, Gotify
Workplace: Pumble, Flock, Twist, DingTalk, Zalo, Webhooks
Each adapter supports per-channel model overrides, DM/group policies, rate limiting, and output formatting.
WhatsApp Web Gateway (QR Code)
Connect your personal WhatsApp account to OpenFang via QR code, just like WhatsApp Web. No Meta Business account required.
Prerequisites
- Node.js >= 18 installed (download)
- OpenFang installed and initialized
Setup
1. Install the gateway dependencies:
cd packages/whatsapp-gateway
npm install
2. Configure config.toml:
[channels.whatsapp]
mode = "web"
default_agent = "assistant"
3. Set the gateway URL (choose one):
Add to your shell profile for persistence:
# macOS / Linux
echo 'export WHATSAPP_WEB_GATEWAY_URL="http://127.0.0.1:3009"' >> ~/.zshrc
source ~/.zshrc
Or set it inline when starting the gateway:
export WHATSAPP_WEB_GATEWAY_URL="http://127.0.0.1:3009"
4. Start the gateway:
node packages/whatsapp-gateway/index.js
The gateway listens on port 3009 by default. Override with WHATSAPP_GATEWAY_PORT.
5. Start OpenFang:
openfang start
# Dashboard at http://localhost:4200
6. Scan the QR code:
Open the dashboard → Channels → WhatsApp. A QR code will appear. Scan it with your phone:
WhatsApp → Settings → Linked Devices → Link a Device
Once scanned, the status changes to connected and incoming messages are routed to your configured agent.
Gateway Environment Variables
| Variable |
Description |
Default |
WHATSAPP_WEB_GATEWAY_URL |
Gateway URL for OpenFang to connect to |
(empty = disabled) |
WHATSAPP_GATEWAY_PORT |
Port the gateway listens on |
3009 |
OPENFANG_URL |
OpenFang API URL the gateway reports to |
http://127.0.0.1:4200 |
OPENFANG_DEFAULT_AGENT |
Agent that handles incoming messages |
assistant |
Gateway API Endpoints
| Method |
Route |
Description |
POST |
/login/start |
Generate QR code (returns base64 PNG) |
GET |
/login/status |
Connection status (disconnected, qr_ready, connected) |
POST |
/message/send |
Send a message ({ "to": "5511999999999", "text": "Hello" }) |
GET |
/health |
Health check |
Alternative: WhatsApp Cloud API
For production workloads, use the WhatsApp Cloud API with a Meta Business account. See the Cloud API configuration docs.
27 LLM Providers, 123+ Models
3 native drivers (Anthropic, Gemini, OpenAI-compatible) route to 27 providers:
Anthropic, Gemini, OpenAI, Groq, DeepSeek, OpenRouter, Together, Mistral, Fireworks, Cohere, Perplexity, xAI, AI21, Cerebras, SambaNova, HuggingFace, Replicate, Ollama, vLLM, LM Studio, Qwen, MiniMax, Zhipu, Moonshot, Qianfan, Bedrock, and more.
Intelligent routing with task complexity scoring, automatic fallback, cost tracking, and per-model pricing.
Migrate from OpenClaw
Already running OpenClaw? One
OpenFang 是一个用 Rust 编写的开源 Agent 操作系统。它不是一个等着你输入提示词的聊天机器人,而是让自主智能体 7×24 小时不间断运行——按计划执行任务、监控目标、构建知识图谱,并把结果汇报到仪表盘上。