A minimal browser automation agent using Google's Gemini 2.5 Computer Use Preview model and Playwright for web browser control.

For browser automation without a sandbox, use this project https://github.com/pmbstyle/gemini-browser-agent

image

• Visual Browser Control: Uses screenshots to "see" and interact with web pages
• Automated Actions: Supports mouse clicks, keyboard input, scrolling, navigation, and more
• Safety Controls: Built-in confirmation prompts for risky actions
• Human-in-the-Loop: Optional user confirmation for sensitive operations

• open_web_browser, navigate, search
• click_at, hover_at, type_text_at
• key_combination, scroll_document, scroll_at
• drag_and_drop, go_back, go_forward
• wait_5_seconds

conda create -n gemcu python=3.11 -y
conda activate gemcu

python -m pip install --upgrade pip
python -m pip install google-genai playwright termcolor

playwright install chromium

# Windows PowerShell
$env:GEMINI_API_KEY="PASTE_YOUR_KEY_HERE"
# Linux/Mac
export GEMINI_API_KEY="PASTE_YOUR_KEY_HERE"

python agent.py "Find Wikipedia article about Niagara Falls and open History section"

• Python 3.11+
• Gemini API key (Get API key)
• Chrome/Chromium browser

This agent runs in a controlled browser environment. For production use, consider running in a sandboxed virtual machine or container for additional security.

Based on Google's Gemini Computer Use API.