A minimal API service that executes browser automation tasks using natural language instructions powered by browser-use and Gemini.
- Natural language browser automation
- Synchronous API (request → execute → response)
- JSON structured results
- Browser instance reuse for efficiency
- Zero configuration (just add your API key)
- Python 3.11+
- Gemini API key (Get one here)
- uv - Modern Python package manager
# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Or via Homebrew
brew install uvcd limbo
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY# Install Python dependencies
uv syncNote: Limbo uses your local Google Chrome installation, so no additional browser installation is needed.
uv run uvicorn main:app --reloadService will start at http://localhost:8000
# Simple test
curl -X POST http://localhost:8000/execute \
-H "Content-Type: application/json" \
-d '{
"instruction": "访问 example.com，返回页面标题"
}'
# Real-world example: Extract Hacker News headlines
curl -X POST http://localhost:8000/execute \
-H "Content-Type: application/json" \
-d '{
"instruction": "访问 news.ycombinator.com，提取前5条新闻的标题和链接，返回JSON数组格式"
}'Execute a browser automation task with natural language instruction.
Request Body:
{
"instruction": "string - Natural language instruction for the browser task",
"return_screenshot": false
}Response:
{
"success": true,
"result": {
"data": "Structured data extracted/processed by the agent"
},
"execution_time": 12.5
}Error Response:
{
"success": false,
"error": "Error message describing what went wrong",
"execution_time": 2.1
}curl -X POST http://localhost:8000/execute \
-H "Content-Type: application/json" \
-d '{
"instruction": "访问 GitHub trending，提取前3个仓库的名称、描述和star数，返回JSON数组"
}'curl -X POST http://localhost:8000/execute \
-H "Content-Type: application/json" \
-d '{
"instruction": "访问 weather.com，搜索北京天气，返回今天的温度和天气状况"
}'curl -X POST http://localhost:8000/execute \
-H "Content-Type: application/json" \
-d '{
"instruction": "访问 httpbin.org/forms/post，填写姓名为张三，提交表单，返回响应结果"
}'Create a .env file in the project root:
# Required
GEMINI_API_KEY=your_gemini_api_key_here
# Optional
BROWSER_HEADLESS=true # Set to false to see browser actions
LOG_LEVEL=INFO # DEBUG, INFO, WARNING, ERRORlimbo/
├── main.py # FastAPI application entry point
├── browser_service.py # Browser-use service wrapper
├── pyproject.toml # Python dependencies
├── .env # Environment configuration (gitignored)
├── .env.example # Environment template
├── docs/
│ └── v1.md # Detailed implementation plan
└── README.md # This file
# Auto-reload on code changes
uv run uvicorn main:app --reload --log-level debugFastAPI provides auto-generated interactive docs:
- Swagger UI: http://localhost:8000/docs
- ReDoc: http://localhost:8000/redoc
Make sure you have Google Chrome installed on your system. Limbo will automatically detect and use your local Chrome installation.
# Verify your .env file exists and has the correct key
cat .env | grep GEMINI_API_KEY# Use a different port
uv run uvicorn main:app --port 8001- Single request at a time (no concurrency)
- No request timeout configuration
- No authentication/authorization
- No task persistence
- Browser stays in memory (restart to clear)
See docs/v1.md for detailed implementation plan and design decisions.
MIT
This is a minimal V1 implementation. For feature requests or improvements, please open an issue first to discuss the changes.