🚀 The First "One-Agent Company" Framework — Let AI handle everything from discovering needs to shipping products.
🌟 Supported LLMs: Claude · DeepSeek · Grok (xAI) · Qwen · Ollama (Llama/Qwen/Hermes/OpenClaw)
🤖 What if an AI could start a company all by itself?
AgentFounder is the world's first open-source framework that turns a single AI agent into a One-Agent Company (OAC). It autonomously discovers real-world problems, assembles a virtual team of specialized AI models, writes and deploys products, collects user feedback, and iteratively improves — without any human coding, prompting, or management.
Just run AgentFounder and watch your AI company come to life.
AgentFounder is a fully automated AI entrepreneurship framework. Launch it, and it will:
- 🔍 Scan the entire web (Reddit, Hacker News, ProductHunt) to capture real user pain points and "I wish there was" demands;
- 🤝 Intelligently assemble a virtual AI team based on needs (automatically match the best LLMs as "co-founders");
- 📝 Auto-generate funding proposals to collect feedback commitments from early adopters;
- 💻 Auto-write and deploy minimum viable products (Telegram Bot / Web App) — zero-code deployment;
- 📊 Continuously collect user feedback, calculate the "Relief Index", and automatically iterate for improvement.
Zero lines of code, zero business plans, zero bank accounts required — just sit back and watch your first AI product come to life.
"You don't need a co-founder. You need AgentFounder." — Zero human employees, infinite execution.
| Traditional Startup | AgentFounder Startup |
|---|---|
| Months of market research | Minutes of web-wide scanning |
| Hunting for co-founders | Auto-match optimal AI models |
| Writing decks, pitching VCs | Auto-generate funding proposals, acquire early adopters |
| Hiring developers, waiting for sprints | Auto-generate code and deploy (with fallback templates) |
| Manual feedback collection | Auto-calculate "Relief Index" to drive iterations |
This is not a "tool that helps you write code" — it's an "AI agent that starts a business for you."
# 1. Clone the project
git clone https://github.com/usaos/AgentFounder.git
cd AgentFounder
# 2. Configure environment (copy and fill in your API Keys)
cp .env.example .env
# Edit .env file, at least fill in TAVILY_API_KEY (for search), etc.
# 3. One-click deploy (auto-pulls models, starts services)
chmod +x scripts/install.sh
./scripts/install.shAfter startup, visit:
- 📊 Dashboard: http://localhost:8000/dashboard
- 📘 API Docs: http://localhost:8000/docs
- 💚 Health Check: http://localhost:8000/health
| Module | Function |
|---|---|
| OpportunityScanner | Scans the web for demands, supports Tavily search + mock fallback, with vector deduplication |
| TeamBuilder | Matches registered AI agents via vector similarity, supports weighted ranking |
| FundingEngine | Auto-generates funding proposal text |
| ProductFactory | Auto-generates Python code, syntax validation, built-in Telegram Bot / Web App templates |
| OperationLoop | Monitors "Relief Index", triggers optimization suggestions on low scores |
| MeaningOracle | Computes Relief Index based on Sentence‑Transformers |
| Mesh Bus | WebSocket message bus, supports real-time collaboration with external Agents |
| Health Monitor | Background health checks for registered agents, auto-removes offline ones |
| Token Monitor | Daily token limit, auto-fuse on overage to prevent unexpected costs |
| Global Auth | Optional global API Token authentication to secure endpoints |
| Dashboard | Gradio panel for visualizing opportunities, testing Relief Index, online API Key configuration |
AgentFounder has built-in task routing that automatically selects the best model based on task type (configurable):
| Task Type | Preferred Model | Fallback |
|---|---|---|
Demand Scan (demand_scan) |
Claude Haiku | Ollama (llama3.1) |
Code Generation (code_gen) |
DeepSeek Coder | Ollama (llama3.1) |
Trend Summary (summary_trend) |
DeepSeek Chat | Ollama (llama3.1) |
Emotion Analysis (emotion_analysis) |
Grok (xAI) / Qwen | Ollama (llama3.1) |
External Agents can register and collaborate via HTTP or WebSocket.
AgentFounder/
├── docker-compose.yml # One-click launch API + Ollama
├── Dockerfile # API image
├── requirements.txt
├── .env.example # Configuration template
├── scripts/
│ ├── install.sh # Linux/Mac one-click install script
│ ├── install.bat # Windows one-click install script
│ └── pull_models.sh # Batch pull local models
├── app/
│ ├── main.py # FastAPI entry
│ ├── scheduler.py # Fully automated scheduled orchestrator
│ ├── api/ # REST + WebSocket routes, authentication
│ ├── core/ # Six core engines
│ ├── agents/ # Agent registry, Mesh bus, health monitoring
│ ├── llm/ # Multi-model adapters + smart routing
│ └── utils/ # Database, encryption, logging, token fusing, etc.
├── dashboard/
│ └── app.py # Gradio control panel
└── templates/ # Product templates (Telegram Bot / Web App)
Key environment variables (.env file):
| Variable | Description | Required |
|---|---|---|
API_GLOBAL_TOKEN |
Global access token (strongly recommended to change) | Recommended |
TAVILY_API_KEY |
Tavily search API Key | Recommended |
ANTHROPIC_API_KEY |
Claude API Key | Optional |
DEEPSEEK_API_KEY |
DeepSeek API Key | Optional |
X_GROK_API_KEY |
Grok (xAI) API Key | Optional |
DASHSCOPE_API_KEY |
Qwen (Tongyi Qianwen) API Key | Optional |
OLLAMA_HOST |
Ollama service URL (keep default for Docker) | Default |
SCAN_INTERVAL_MIN |
Full-automation scan interval (minutes) | Default 30 |
DAILY_TOKEN_LIMIT |
Daily token fuse limit | Default 100000 |
VERCEL_TOKEN |
Optional, for auto-deployment to Vercel | Optional |
All API Keys are automatically encrypted with AES and will not be exposed in logs.
A[Scan Web for Needs] --> B[Vector Dedup]
B --> C[Assemble AI Team]
C --> D[Generate Funding Proposal]
D --> E[Auto-Write Code]
E --> F[Deploy Product]
F --> G[Collect User Feedback]
G --> H[Calculate "Relief Index"]
H -->|Low Score| A
The entire process is triggered by APScheduler, running once every 30 minutes by default, with no manual intervention required.
- ✅ AGPL-3.0 open-source license, ensuring community sharing.
- ✅ AES encryption for all API keys, never leaked in logs.
- ✅ Global API Token authentication (optional) to prevent unauthorized access.
- ✅ Daily token fusing to prevent unexpected overages.
- ✅ Crawler rate limiting (5-second cooldown) to reduce ban risks.
- ✅ Periodic data archiving to prevent database bloat.
We welcome any form of contribution (code, documentation, ideas). Please read CONTRIBUTING.md for guidelines.
- Fork this project;
- Create your feature branch (
git checkout -b feature/AmazingFeature); - Commit your changes (
git commit -m 'Add some AmazingFeature'); - Push to the branch (
git push origin feature/AmazingFeature); - Open a Pull Request.
This project is licensed under AGPL-3.0.
- 🐦 Follow us on X (coming soon)
- 💬 Join our [Discord community] (coming soon)
- 📧233961878@qq.com
Thanks to all open-source communities and LLM providers that made this project possible. Special thanks to Anthropic, DeepSeek, xAI, Alibaba Cloud, Ollama, and the Sentence‑Transformers project.
If you like this project, please give it a Star ⭐ to help more people discover it!
🍴 Fork and contribute — we welcome PRs!
🐦 Share your OAC journey on X with #AgentFounder and #OneAgentCompany
AgentFounder — Because the best co-founder is code. 🚀