Jeomon/Web-Use

What it solves

Web-Use is an autonomous browsing agent that automates complex web tasks—such as filling forms, performing searches, and interacting with dynamic content—reducing the need for manual browser navigation.

How it works

The agent uses Large Language Models (LLMs) and the Chrome DevTools Protocol (CDP) to interact with the web. It builds a "semantic tree" of the page's DOM to understand the structure and roles of elements (like buttons and inputs) without relying on XPaths. It also supports vision capabilities for visual understanding and a custom protocol called WebMCP, which allows websites to expose specific tools directly to the agent.

Who it’s for

Developers and automation engineers who want to integrate intelligent, LLM-powered web navigation and interaction into their workflows.

Highlights

• Multi-LLM Support: Compatible with providers like Anthropic Claude, Google Gemini, OpenAI, Groq, and Ollama.
• OAuth 2.0 + PKCE: Built-in support for authenticated workflows, allowing the agent to log into services like Google or GitHub without storing passwords.
• Semantic Tree: Uses real DOM parent-child relationships to provide the agent with accurate structural context.
• WebMCP: Automatically discovers and uses custom tools exposed by supported websites.
• State Awareness: Maintains understanding of the page state to avoid loops and errors.
• File Operations: Capable of downloading files and uploading content to forms.