Summary

AgentChat provides a Gemini Web Automation system using Chrome CDP and Playwright, enabling "thinker + executor" synergy between planning LLMs (e.g., DeepSeek) and reasoning LLMs (Gemini Pro). It grants multimodal web interaction capabilities—screenshot analysis, browsing, UI control—to any LLM, offering a near-zero-cost solution for complex AI agent development and web automation.

How It Works

The system orchestrates LLM collaboration via Chrome CDP. A planning LLM formulates strategies, which are injected into Gemini Web. Playwright automates browser actions, allowing Gemini to analyze screenshots, browse pages, and interact with UIs. This leverages the free Gemini Web interface for inference, drastically reducing costs compared to API-based solutions while equipping LLMs with multimodal web access.

Quick Start & Requirements

• Installation: Clone repo, install playwright and websocket-client, run python3 -m playwright install chromium, configure .env (especially PROXY_SERVER), and execute scripts/setup.sh.
• Execution: Launch Chrome/Gemini with scripts/start-chrome-debug.sh and connect using scripts/connect-gemini.sh.
• Prerequisites: Python 3.8+, Playwright, Chromium, websocket-client, and a functional HTTP/SOCKS5 proxy (essential for mainland China). Google account recommended for Gemini Web Pro.
• Setup: Estimated at 5 minutes.

Highlighted Details

• Cost-Effective Reasoning: Leverages free Gemini Web for inference, making deep reasoning nearly cost-free.
• Multimodal LLM Empowerment: Equips text LLMs with advanced web interaction, including visual analysis and UI control.
• Network Environment Solution: Includes Chrome flags to overcome GFW issues in mainland China.
• Extensible Framework: Designed as a CDP bridge for any LLM to any web application.

Maintenance & Community

No specific details on maintainers, community channels, or roadmap were found in the provided README.

Licensing & Compatibility

• License: MIT.
• Compatibility: MIT license permits commercial use and integration into closed-source projects.

Limitations & Caveats

• Proxy Dependency: HTTP/SOCKS5 proxy is mandatory for mainland China users due to GFW interference; VLESS Reality proxies are incompatible.
• Gemini Pro Access: Requires Google account login for advanced features.
• UI Input Workaround: A specific input method (type comma, paste, delete comma) is noted for potential Gemini UI interaction issues.
• Chrome Startup Resilience: Relies on injected flags to mitigate Chrome's fail-safe mode triggered by network restrictions.