webbrain-one/webbrain

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

What it solves

WebBrain provides a way to interact with the web using an AI agent that can read, navigate, and perform actions on web pages. It eliminates the need for manual repetitive tasks by allowing users to automate browser workflows, chat with the content of a page, and schedule recurring web-based tasks.

How it works

It is a browser extension (available for Chrome, Firefox, and Edge) that integrates an AI agent into a side panel. Instead of relying on brittle CSS selectors, it uses the accessibility tree to understand page elements. Users can connect it to various LLMs, including local servers (via llama.cpp or Ollama), cloud APIs (OpenAI, Anthropic, etc.), or a managed default provider. The agent operates in three modes: "Ask" for read-only queries, "Act" for performing actions like clicking and typing, and "Dev" for technical page diagnostics.

Who it’s for

WebBrain is designed for users who want to automate web-based workflows, researchers who need to extract information from multiple pages, and developers who want an AI-powered assistant to help them navigate and interact with web applications.

Highlights

  • Flexible Model Support: Compatible with over 100 cloud providers and local LLM servers.
  • Autonomous Tool-Use: Capable of running multi-step loops (up to 195 steps) to complete complex tasks.
  • Saved Workflows: Ability to turn successful agent runs into reusable and shareable workflows.
  • Scheduled Tasks: Support for /schedule and /watch commands to poll pages and act when specific conditions are met.
  • Smart Context Management: Features token-aware auto-compaction to handle long conversations within model context limits.