openfang-agent-os

What this skill does

```markdown
---
name: openfang-agent-os
description: Expertise in OpenFang, the open-source Agent Operating System built in Rust for running autonomous AI agents on schedules with Hands, tools, and MCP support.
triggers:
  - set up OpenFang agent operating system
  - create an autonomous agent with OpenFang
  - configure OpenFang Hands
  - build a custom Hand for OpenFang
  - OpenFang CLI commands and usage
  - integrate LLM providers with OpenFang
  - OpenFang HAND.toml configuration
  - deploy OpenFang agents on a schedule
---

# OpenFang Agent OS

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

OpenFang is an open-source **Agent Operating System** written in Rust that runs autonomous AI agents 24/7 without requiring user prompts. Unlike chatbot frameworks, OpenFang agents work *for* you — on schedules, building knowledge graphs, monitoring targets, generating leads, and delivering results to a dashboard. The entire system compiles to a single ~32MB binary.

---

## Installation

### Linux / macOS
```bash
curl -fsSL https://openfang.sh/install | sh
```

### Windows (PowerShell)
```powershell
irm https://openfang.sh/install.ps1 | iex
```

### From source (requires Rust toolchain)
```bash
git clone https://github.com/RightNow-AI/openfang.git
cd openfang
cargo build --release
# Binary at ./target/release/openfang
```

### Initialize and start
```bash
openfang init          # Scaffolds config in ~/.openfang/
openfang start         # Starts daemon + dashboard at http://localhost:4200
```

---

## Core Concepts

| Concept | Description |
|---|---|
| **Hand** | Autonomous capability package (researcher, lead gen, OSINT, etc.) |
| **HAND.toml** | Manifest defining tools, schedule, settings, and metrics for a Hand |
| **SKILL.md** | Domain expertise injected into agent context at runtime |
| **FangHub** | Registry for sharing/installing community Hands |
| **MCP** | Model Context Protocol — tool/resource interface for agents |
| **Channel Adapter** | Integration layer (Telegram, Slack, WhatsApp, webhooks, etc.) |

---

## CLI Reference

### Global Commands
```bash
openfang init                    # Initialize config and workspace
openfang start                   # Start the OpenFang daemon
openfang stop                    # Stop the daemon
openfang status                  # Show daemon and agent status
openfang logs                    # Tail all logs
openfang logs --hand researcher  # Tail logs for a specific Hand
openfang upgrade                 # Upgrade to latest version
```

### Hand Management
```bash
openfang hand list               # List all available Hands
openfang hand activate <name>    # Activate a Hand (starts running immediately)
openfang hand pause <name>       # Pause without losing state
openfang hand resume <name>      # Resume from paused state
openfang hand stop <name>        # Stop and clear state
openfang hand status <name>      # Show Hand run status and metrics
openfang hand inspect <name>     # Show full Hand manifest and config
openfang hand run <name>         # Trigger a one-off manual run
openfang hand logs <name>        # Stream Hand-specific logs
```

### FangHub (Community Registry)
```bash
openfang hub search "crypto monitor"   # Search available Hands
openfang hub install <author>/<hand>   # Install a community Hand
openfang hub publish                   # Publish your Hand to FangHub
openfang hub update <hand>             # Update an installed Hand
```

### Config Management
```bash
openfang config show                   # Print current config
openfang config set llm.provider openai
openfang config set llm.model gpt-4o
openfang config get llm.provider
```

### Provider / Key Management
```bash
openfang provider list                 # List configured LLM providers
openfang provider add openai           # Interactively add a provider
openfang provider set-default openai   # Set default provider
```

---

## Configuration

OpenFang reads from `~/.openfang/config.toml` (created by `openfang init`).

```toml
# ~/.openfang/config.toml

[core]
workspace = "~/.openfang/workspace"
log_level  = "info"          # trace | debug | info | warn | error
dashboard_port = 4200
daemon_port    = 7700

[llm]
provider = "openai"          # openai | anthropic | gemini | groq | ollama | ...
model    = "gpt-4o"
temperature = 0.7
max_tokens  = 8192
timeout_secs = 120

[llm.fallback]               # Automatic fallback chain
providers = ["anthropic", "groq"]

[security]
approval_required = ["browser.purchase", "twitter.post", "email.send"]
sandbox_mode = false         # true = no outbound network in tools
audit_log = true

[scheduler]
timezone = "America/New_York"
max_concurrent_hands = 4

[channels.telegram]
bot_token = "${TELEGRAM_BOT_TOKEN}"   # Always use env var references
chat_id   = "${TELEGRAM_CHAT_ID}"

[channels.slack]
webhook_url = "${SLACK_WEBHOOK_URL}"

[storage]
backend = "sqlite"           # sqlite | postgres
sqlite_path = "~/.openfang/db/openfang.db"
# postgres_url = "${DATABASE_URL}"
```

### Environment Variables
```bash
# LLM Providers
export OPENAI_API_KEY="${OPENAI_API_KEY}"
export ANTHROPIC_API_KEY="${ANTHROPIC_API_KEY}"
export GEMINI_API_KEY="${GEMINI_API_KEY}"
export GROQ_API_KEY="${GROQ_API_KEY}"

# Channels
export TELEGRAM_BOT_TOKEN="${TELEGRAM_BOT_TOKEN}"
export TELEGRAM_CHAT_ID="${TELEGRAM_CHAT_ID}"
export SLACK_WEBHOOK_URL="${SLACK_WEBHOOK_URL}"

# Storage (if using Postgres)
export DATABASE_URL="${DATABASE_URL}"
```

---

## Bundled Hands — Activation & Config

### Researcher Hand
```bash
openfang hand activate researcher
```

Configure in `~/.openfang/hands/researcher/config.toml`:
```toml
[researcher]
schedule     = "0 7 * * *"          # Cron: run at 7 AM daily
topic        = "Rust async runtimes 2026"
depth        = "deep"               # quick | standard | deep
sources      = ["web", "arxiv", "news"]
output_format = "markdown"          # markdown | pdf | json
language     = "en"
deliver_to   = ["telegram", "dashboard"]
cite_sources = true
credibility_check = true            # Uses CRAAP criteria
```

### Lead Hand
```bash
openfang hand activate lead
```

```toml
[lead]
schedule = "0 6 * * 1-5"           # Weekdays at 6 AM
icp = """
  B2B SaaS companies, 10-200 employees,
  Series A or B, hiring engineers,
  US or Canada
"""
score_threshold = 65                # Only deliver leads scored >= 65
deduplicate = true
output_format = ["csv", "json"]
deliver_to = ["telegram", "dashboard"]
max_leads_per_run = 50
```

### Collector Hand (OSINT)
```bash
openfang hand activate collector
```

```toml
[collector]
schedule  = "*/30 * * * *"         # Every 30 minutes
targets   = [
  { type = "company", value = "Acme Corp" },
  { type = "topic",   value = "AI regulation EU" },
]
alert_on_change = true
sentiment_tracking = true
knowledge_graph   = true
deliver_to = ["telegram"]
alert_threshold = "significant"    # minor | significant | critical
```

### Twitter Hand
```bash
openfang hand activate twitter
```

```toml
[twitter]
schedule        = "0 9,13,18 * * *"  # Post 3x daily
account_handle  = "@yourhandle"
topics          = ["Rust", "AI agents", "open source"]
content_formats = ["thread", "single", "poll", "question"]
approval_queue  = true              # Posts queue for your approval
respond_to_mentions = true
track_metrics   = true
```

### Browser Hand
```bash
openfang hand activate browser
```

```toml
[browser]
# NOTE: purchase_approval is always enforced regardless of config
headless        = true
session_persist = true
playwright_port = 9222
allowed_domains = ["*"]            # Restrict with ["example.com"]
```

---

## Building a Custom Hand

### Directory structure
```
my-hand/
├── HAND.toml       # Required: manifest
├── SKILL.md        # Required: domain expertise injected at runtime
├── prompts/
│   └── system.md   # Required: multi-phase operational playbook
└── tools/
    └── custom.toml # Optional: additional tool definitions
```

### HAND.toml
```toml
[hand]
name        = "price-monitor"
ver