AGENT OPERATING SYSTEM

The Agent
Operating System

Open-source Agent OS built in Rust. 7 autonomous Hands that run on schedules, build knowledge graphs, and report to your dashboard. 30 agents, 40 channels, 38 tools, 26 LLM providers. 16 security systems. One binary.

14 crates. 137K lines of Rust. Zero clippy warnings. Production-ready.

>curl -fsSL https://openfang.sh/install | sh
0
Hands
0
Security Systems
0+
Channels
0
LLM Providers
FEATURES

What OpenFang Does

The core primitives for building, running, and deploying autonomous agents.

HANDS

7 Autonomous Capability Packages

Pre-built agents that work FOR you. Clip turns video into shorts. Lead generates qualified leads. Collector monitors targets. Predictor forecasts with Brier scores. Researcher fact-checks with CRAAP. Twitter manages your X account. Browser automates the web. Activate, configure, check your dashboard.

SECURITY

16 Security Systems

WASM dual-metered sandbox, Ed25519 manifest signing, Merkle audit trail, taint tracking, SSRF protection, secret zeroization, HMAC-SHA256 mutual auth, GCRA rate limiter, subprocess isolation, prompt injection scanner, path traversal prevention, and more.

RUNTIME

Sandboxed Execution

Tool code runs inside WASM with dual metering (fuel + epoch interruption). File operations are workspace-confined. Subprocesses are env-cleared and timeout-enforced. 10-phase graceful shutdown.

AGENTS

30 Pre-Built Agents

From orchestrators to code reviewers to customer support. Four performance tiers across Anthropic, Gemini, Groq, and DeepSeek. Spawn one in a single command.

TOOLS

38 Built-In Tools + MCP

38 native tools plus Model Context Protocol client and server. Connect external MCP servers and expose OpenFang tools to other agents. Includes web search, browser automation, image generation, TTS, Docker, knowledge graphs.

MEMORY

Persistent Memory

SQLite-backed storage with vector embeddings. Cross-channel canonical sessions, automatic LLM-based compaction, and JSONL session mirroring. Agents recall context across conversations and channels.

CHANNELS

40 Channel Adapters

Telegram, Discord, Slack, WhatsApp, Teams, IRC, Matrix, and 33 more. Per-channel model overrides, DM/group policies, rate limiting, and output formatting. One agent, every platform.

PROTOCOLS

MCP + A2A + OFP

Model Context Protocol (client + server), Google A2A agent-to-agent tasks, and OpenFang Protocol for P2P networking with HMAC-SHA256 mutual authentication. Your agents talk to the world.

DESKTOP

Tauri 2.0 Native App

Native desktop application with system tray, notifications, single-instance enforcement, auto-start on login, and global shortcuts. Full dashboard in a native window.

COMPARE

OpenFang vs The Landscape

6 frameworks. 10 dimensions. See how OpenFang's kernel-grade Rust architecture compares to OpenClaw, ZeroClaw, CrewAI, AutoGen, and LangGraph.

$openfang compare --all
[10 features × 6 frameworks]
Feature
OpenFang
OpenClaw
ZeroClaw
CrewAI
AutoGen
LangGraph
01Language
Rust
TypeScript
Rust
Python
Python
Python
02Autonomous Hands
7 built-in
None
None
None
None
None
03Security Layers
16 discrete
3 basic
6 layers
1 basic
Docker
AES enc.
04Agent Sandbox
WASM dual
None
Allowlists
None
Docker
None
05Channel Adapters
40
13
15
0
0
0
06Built-in Tools
53 + MCP
50+
12
Plugins
MCP
LC tools
07Memory
SQLite + vec
File-based
SQLite FTS5
4-layer
External
Checkpoints
08Desktop App
Tauri 2.0
None
None
None
Studio
None
09Audit Trail
Merkle chain
Logs
Logs
Tracing
Logs
Checkpoints
10License
MIT
MIT
MIT
MIT
Apache 2.0
MIT
OpenFang 9
ZeroClaw 1
// feature-by-feature comparison
BENCHMARKS

Measured, Not Marketed

Cold start time, memory footprint, install size, security depth, channel coverage, and provider support — benchmarked across 6 leading agent frameworks.

OpenFang
OpenClaw
ZeroClaw
CrewAI
AutoGen
LangGraph
Cold Start Timelower is better
ZeroClaw
10ms
OpenFang
180 ms
LangGraph
2500 ms
CrewAI
3000 ms
AutoGen
4000 ms
OpenClaw
5980 ms
Idle Memory Usagelower is better
ZeroClaw
5 MB
OpenFang
40 MB
LangGraph
180 MB
CrewAI
200 MB
AutoGen
250 MB
OpenClaw
394 MB
Install Sizelower is better
ZeroClaw
8.8 MB
OpenFang
32 MB
CrewAI
100 MB
LangGraph
150 MB
AutoGen
200 MB
OpenClaw
500 MB
Security Systemshigher is better
OpenFang
16 layers
ZeroClaw
6 layers
OpenClaw
3 layers
AutoGen
2 layers
LangGraph
2 layers
CrewAI
1 layers
Channel Adaptershigher is better
OpenFang
40 built-in
ZeroClaw
15 built-in
OpenClaw
13 built-in
CrewAI
0
AutoGen
0
LangGraph
0
LLM Providershigher is better
ZeroClaw
28 native
OpenFang
27 native
LangGraph
15 native
CrewAI
10 native
OpenClaw
10 native
AutoGen
8 native
// benchmarks from official docs and public repos — February 2026
HANDS

Agents That Work For You

Hands are pre-built autonomous capability packages. Unlike regular agents you chat with, Hands run on schedules, build knowledge graphs, and report results to your dashboard. Activate one and check in on its progress.

>

Traditional agents wait for you to type. Hands work for you.

Each Hand bundles a HAND.toml manifest, system prompt with multi-phase operational playbook, SKILL.md expert knowledge, configurable settings, and dashboard metrics — all compiled into the binary at build time.

🎬

Clip

Content

Turns long-form video into viral short clips with captions, thumbnails, and optional AI voice-overs. Publishes to Telegram and WhatsApp.

8-phase pipelineFFmpeg + yt-dlp5 STT backendsAuto-publish
📊

Lead

Data

Autonomous lead generation. Discovers, enriches, scores, and deduplicates qualified leads on a daily schedule. Builds ICP profiles and knowledge graphs.

ICP scoring 0-100Web research loopsCSV/JSON/Markdown exportScheduled delivery
🔍

Collector

Intelligence

OSINT-style intelligence collector. Monitors any target continuously with change detection, sentiment tracking, and knowledge graph construction.

Change detectionSentiment analysisKnowledge graphsCritical alerts
🔮

Predictor

Forecasting

Superforecasting engine. Collects signals, builds calibrated reasoning chains, makes predictions, and tracks accuracy with Brier scores.

Brier score calibrationContrarian modeEvidence chainsAccuracy tracking
🔬

Researcher

Productivity

Deep autonomous researcher. Cross-references sources, fact-checks claims with CRAAP evaluation, and generates cited reports in multiple languages.

CRAAP fact-checkingMulti-languageAPA citationsSource verification
𝕏

Twitter

Communication

Autonomous Twitter/X manager. Creates content in 7 rotating formats, schedules posts, engages with mentions, and tracks performance metrics.

Approval queue7 content typesEngagement trackingBrand voice control
🌐

Browser

Automation

Web automation agent. Navigates sites, fills forms, clicks buttons, and completes multi-step workflows. Mandatory purchase approval gate.

Purchase gatePlaywright bridgeSession persistenceCAPTCHA detection
+

Build Your Own

Define a HAND.toml with tools, settings, requirements, and a system prompt. Publish to FangHub.

Hand development guide →
$openfang hand --help
activate <hand> Activate a Hand (spawns autonomous agent)
deactivate <hand> Stop a running Hand
status <hand> Check Hand metrics and status
list List all available Hands
pause <hand> Pause a running Hand
resume <hand> Resume a paused Hand
Built & maintained byRightNow AIRightNow

OpenFang is built by Jaber, Founder of RightNow.