ContextQMD
Back to UI-TARS-desktop

Open Operator

examples/operator-browserbase/README.md

0.3.03.8 KB
Original Source

Open Operator

Fork from browserbase/open-operator

[!WARNING] This is simply a proof of concept. Browserbase aims not to compete with web agents, but rather to provide all the necessary tools for anybody to build their own web agent. We strongly recommend you check out both Browserbase and our open source project Stagehand to build your own web agent.

https://github.com/user-attachments/assets/354c3b8b-681f-4ad0-9ab9-365dbde894af

Getting Started

First, install the dependencies for this repository. This requires pnpm.

<!-- This doesn't work with NPM, haven't tested with yarn -->
bash
pnpm install

Next, copy the example environment variables:

bash
cp .env.example .env.local

You'll need to set up your API keys:

  1. Get your UI-TARS Service from UI-TARS
  2. Get your Browserbase API key and project ID from Browserbase

Update .env.local with your API keys:

  • UI_TARS_BASE_URL: Your UI-TARS Base Url
  • UI_TARS_API_KEY: Your UI-TARS API Key
  • UI_TARS_MODEL: Your UI-TARS Model
  • BROWSERBASE_API_KEY: Your Browserbase API key
  • BROWSERBASE_PROJECT_ID: Your Browserbase project ID

Then, run the development server:

<!-- This doesn't work with NPM, haven't tested with yarn -->
bash
pnpm dev

Open http://localhost:3000 with your browser to see Open Operator in action.

How It Works

Building a web agent is a complex task. You need to understand the user's intent, convert it into headless browser operations, and execute actions, each of which can be incredibly complex on their own.

Stagehand is a tool that helps you build web agents. It allows you to convert natural language into headless browser operations, execute actions on the browser, and extract results back into structured data.

Under the hood, we have a very simple agent loop that just calls Stagehand to convert the user's intent into headless browser operations, and then calls Browserbase to execute those operations.

Stagehand uses Browserbase to execute actions on the browser, and OpenAI to understand the user's intent.

For more on this, check out the code at this commit.

Key Technologies

  • Browserbase: Powers the core browser automation and interaction capabilities
  • Stagehand: Handles precise DOM manipulation and state management
  • Next.js: Provides the modern web framework foundation
  • OpenAI: Enable natural language understanding and decision making

Contributing

We welcome contributions! Whether it's:

  • Adding new features
  • Improving documentation
  • Reporting bugs
  • Suggesting enhancements

Please feel free to open issues and pull requests.

License

Open Operator is open source software licensed under the MIT license.

Acknowledgments

This project is inspired by OpenAI's Operator feature and builds upon various open source technologies including Next.js, React, Browserbase, and Stagehand.