Introduction
Computer Use Preview is a browser-automation example agent that demonstrates how to use the Gemini Developer API or Vertex AI to execute natural-language instructions and perform multi-step interactions in a real browser. The project provides reproducible examples for research and prototyping, helping developers understand how agents can drive browsers to perform searches, fill forms, capture screenshots, and support visual debugging.
Key Features
- Supports running against Gemini Developer API or Vertex AI backends.
- Uses Playwright for browser automation with support for initial URL injection and interaction scripts.
- Includes example scenarios and test suites for quick onboarding and extension.
Use Cases
- Prototyping automated web tasks and simple data collection workflows.
- Educational and research demonstrations of multi-step agent/browser interactions.
- Evaluating LLM-driven browser operation reliability and explainability in controlled environments.
Technical Highlights
- Built with Python and Playwright, runnable locally or in CI.
- Configurable via environment variables and packaged with example scripts and tests.
- Licensed under Apache-2.0 for research and commercial reuse and extension.