</>MCP Agents Market
Agent

ClawWork

by HKUDS8.5kPythonUpdated 2026-03-03

"ClawWork: OpenClaw as Your AI Coworker - 💰 $15K earned in 11 Hours"

ClawWork is an AI agent framework that transforms language models into economically accountable AI coworkers capable of completing real-world professional tasks. Built on top of Nanobot, it evaluates AI agents across 220 professional tasks from the GDPVal benchmark spanning 44 occupations, requiring agents to manage their own token budgets, earn income by delivering quality work, and maintain economic solvency. The system provides comprehensive tooling for document creation, code execution, web search, and video generation, along with a real-time React dashboard for monitoring agent performance, earnings, and survival metrics. Developers can run agents in standalone simulation mode or integrate ClawWork into existing Nanobot instances to add economic tracking to any AI assistant.

Key Features

Economic survival benchmark where agents start with $10 and must earn income by completing professional tasks while paying for their own API token usage
220 real-world tasks from GDPVal dataset across 44 occupations in manufacturing, finance, healthcare, legal, and technology sectors
8 productivity tools including document creation (Word, Excel, PDF), code sandbox execution, web search, and video generation
LLM-based quality evaluation system with category-specific rubrics and payment based on quality score × BLS hourly wage × estimated hours
Real-time WebSocket dashboard showing balance charts, task completions, learning progress, and economic metrics
ClawMode integration that transforms any Nanobot gateway into an economically accountable coworker across 9 messaging platforms
Multi-agent competition arena supporting GPT-4, Claude, Gemini, Qwen, GLM, and other models with performance leaderboards
Strategic work-versus-learn decision system where agents balance immediate income against long-term knowledge investment

Use Cases

  • 01Benchmarking different LLM models on real-world economic productivity and cost efficiency across professional domains
  • 02Research into AI agent economic viability and autonomous decision-making under resource constraints
  • 03Converting existing Nanobot AI assistants into economically self-sustaining coworkers that must prove ROI
  • 04Evaluating AI readiness for professional work in specific industries like finance, healthcare, or manufacturing
  • 05Testing agent strategies for balancing exploration (learning) versus exploitation (immediate work) in economic environments
  • 06Running multi-agent competitions to identify the most cost-effective AI worker for production deployment

Related Agents

View more

ClawWork — FAQ

What is ClawWork?+

ClawWork is an AI agent framework that evaluates language models as economically accountable coworkers by requiring them to complete real professional tasks from 44 occupations, manage their own API costs, and earn income based on work quality. It transforms AI assistants into agents that must prove economic viability through actual productivity.

How do I install ClawWork?+

Clone the GitHub repository, create a Python 3.10+ environment, install dependencies with pip install -r requirements.txt, copy .env.example to .env and add your OPENAI_API_KEY, then run ./start_dashboard.sh in one terminal and ./run_test_agent.sh in another. The dashboard will be available at http://localhost:3000.

What API keys are required to run ClawWork?+

OPENAI_API_KEY is required for agent operation and task evaluation. E2B_API_KEY is needed if using the default E2B code sandbox (or use CODE_SANDBOX_PROVIDER=boxlite for local execution). WEB_SEARCH_API_KEY (Tavily or Jina) is optional for web search functionality.

Which AI clients does ClawWork work with?+

ClawWork runs as a standalone agent framework in simulation mode. Through ClawMode integration with Nanobot, it supports Telegram, Discord, Slack, WhatsApp, Email, Feishu, DingTalk, MoChat, and QQ as messaging interfaces. It supports any LangChain/LiteLLM-compatible model provider.

Is ClawWork free to use?+

The ClawWork framework itself is open source under MIT license. However, running agents incurs costs from API providers (OpenAI for evaluation and agent inference, E2B for code sandboxes, Tavily/Jina for search) based on your usage. The economic simulation tracks these real costs.

What does the economic survival benchmark measure?+

Agents are evaluated on survival days (economic solvency), work quality (0-1 score), profit margin, token efficiency (income per dollar spent), and task completion rate. Top agents achieve $1,500+/hour equivalent earnings, but must balance income against API token costs that are deducted from their balance in real time.

How do I install ClawWork?+

Open the source repository on GitHub and follow its README. ClawWork is a agent — MCP Agents Market links you directly to the official repo.

Is ClawWork free?+

ClawWork is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches