Skip to main content

​
What Are Hands?

Hands are specialized autonomous agents in OpenFang designed to run continuously and handle specific workflows. Unlike general-purpose chat agents, Hands are optimized for:
  • Continuous operation — Run scheduled tasks 24/7
  • Domain expertise — Deep knowledge in specific areas (video editing, research, social media)
  • Multi-phase pipelines — Complex workflows broken into manageable steps
  • State persistence — Remember progress across sessions
  • Tool specialization — Access to specialized tools for their domain

​
Available Hands

Clip Hand

Turn long videos into viral short clips with captions and thumbnails

Lead Hand

Discover and enrich qualified leads on autopilot

Collector Hand

Monitor any target with continuous intelligence collection

Predictor Hand

Make calibrated predictions and track accuracy over time

Researcher Hand

Conduct exhaustive investigations with fact-checking

Twitter Hand

Autonomous Twitter content creation and engagement

Browser Hand

Navigate websites and complete multi-step web tasks

Custom Hand

Build your own specialized Hand

​
How Hands Work

​
Activation

Hands are activated by loading their definition file: 
openfang hand activate clip
This creates an agent instance with the Hand’s specialized configuration, tools, and system prompt.

​
Configuration

Each Hand has settings that control its behavior: 
# Example from Clip Hand
[settings]
stt_provider = "groq_whisper"  # Speech-to-text provider
publish_target = "telegram"     # Where to send clips
Settings can be configured through:
  • The web dashboard at http://localhost:4200
  • Direct TOML editing in ~/.openfang/hands/
  • CLI configuration commands

​
Requirements

Some Hands require external tools or API keys:
Check the requirements section for each Hand before activating. Missing dependencies will prevent the Hand from functioning.

​
Scheduling

Hands can run on schedules: 
// Created automatically by the Hand's initialization
schedule_create({
  frequency: "daily",
  time: "09:00",
  task: "generate_lead_report"
})

​
Memory & State

Hands persist state between runs:
  • Memory store — Key-value storage for metrics and state
  • Knowledge graph — Entity and relationship tracking
  • File storage — Reports, databases, and artifacts

​
Hand Architecture

Every Hand follows a consistent multi-phase pipeline:
1

Phase 0: Platform Detection

Detect OS and load state from previous runs
2

Phase 1: Initialization

Set up schedules, load configuration, build strategy
3

Phase 2-5: Core Work

Execute the Hand’s specialized workflow (varies by Hand)
4

Final Phase: Persistence

Save state, update metrics, generate reports

​
Dashboard Metrics

All Hands expose metrics through the dashboard:
  • Jobs completed — Total tasks finished
  • Items processed — Clips generated, leads found, etc.
  • Performance data — Success rates, quality metrics
  • Last run — When the Hand last executed

​
Safety Guardrails

Hands include built-in safety features:
  • Rate limiting — Respect API and service limits
  • Approval modes — Require user confirmation for sensitive actions
  • Error recovery — Graceful handling of failures
  • Cross-platform compatibility — Works on Windows, macOS, Linux

​
Choosing the Right Hand

Use CaseHandWhy
Create social media content from videosClip HandAutomated video processing pipeline
Generate sales leadsLead HandContinuous discovery and enrichment
Track competitors or marketsCollector HandChange detection and knowledge graphs
Forecast trendsPredictor HandCalibrated predictions with accuracy tracking
Deep dive researchResearcher HandMulti-source verification and synthesis
Manage Twitter presenceTwitter HandContent generation and engagement
Automate web tasksBrowser HandReal browser automation with safety
Custom workflowCustom HandBuild your own specialized agent

​
Next Steps

Explore Hands

Browse the documentation for each Hand

Build Your Own

Learn how to create custom Hands