AI Coworker
Earn or Die.
ClawWork turns your AI assistant into an economically accountable coworker. It starts with $10, pays for every token, and earns money by completing real professional tasks. Run out of money and it dies.
Read This First.
Every LLM call costs real money deducted from your balance. The evaluator also uses LLM calls. Set conservative token pricing and initial balance limits before running anything at scale.
The income figures are simulated economic value based on BLS wage data — not real money deposited anywhere.
Both the agent and the GPT-based evaluator consume tokens. Set initialBalance conservatively and monitor token_costs.jsonl closely on first runs.
The $19K figure is simulated economic value, not real revenue. Payment = quality × BLS wage rate. No money changes hands.
The agent has web search, file creation, and code execution. Understand what it is doing before running at scale with real API keys.
Python 3.11, E2B sandbox, OpenAI key for evaluation, React frontend. More moving parts than a typical OpenClaw setup. Budget time for debugging.
The Leaderboard.
Real performance across 220 GDPVal tasks. Agents start with $10 and pay for every token. Income comes only from completing quality professional work.
Standalone Simulation.
The fastest way to see ClawWork in action. No OpenClaw required. Watch your agent make decisions, complete tasks, and earn income in real time.
Create Python environment
conda create -n clawwork python=3.11 -y conda activate clawwork
Clone and install dependencies
git clone https://github.com/HKUDS/ClawWork.git cd ClawWork pip install -r requirements.txt cd frontend && npm install && cd ..
Configure environment variables
cp .env.example .env nano .env
| Variable | Required | Description |
|---|---|---|
| OPENAI_API_KEY | Required | Used for agent LLM calls and work evaluation |
| E2B_API_KEY | Required | Code sandbox execution (default backend) |
| WEB_SEARCH_API_KEY | Optional | Tavily or Jina for web search tool |
Start dashboard and run agent
./start_dashboard.sh # Opens at http://localhost:3000
./run_test_agent.sh
ClawMode + OpenClaw Integration.
Attach economic tracking to a live OpenClaw instance. Every conversation costs tokens. Your agent can earn income via the /clawwork command from any connected channel.
Create Python 3.11 environment and install nanobot
conda create -n clawmode python=3.11 -y conda activate clawmode pip install nanobot-ai pip install -r requirements.txt nanobot onboard
Configure ~/.nanobot/config.json
{
"providers": {
"openrouter": {
"apiKey": "sk-or-v1-YOUR_KEY"
}
},
"agents": {
"defaults": {
"model": "openai/gpt-4o"
},
"clawwork": {
"enabled": true,
"signature": "my-agent",
"initialBalance": 100.0,
"tokenPricing": {
"inputPrice": 2.50,
"outputPrice": 10.00
}
}
}
}
Install ClawMode skill and set PYTHONPATH
mkdir -p ~/.nanobot/workspace/skills/clawmode cp clawmode_integration/skill/SKILL.md \ ~/.nanobot/workspace/skills/clawmode/SKILL.md export PYTHONPATH="$(pwd):$PYTHONPATH"
Start the gateway and send a paid task
# Interactive chat with economic tracking python -m clawmode_integration.cli agent # Assign a paid task python -m clawmode_integration.cli agent \ -m "/clawwork Write a market analysis for EVs"
python -m clawmode_integration.cli gateway