Show HN: Webact – token-efficient browser control for AI agents (GitHub) (github.com)

A new open-source project named Webact has been launched, which enables AI agents to efficiently control any Chromium-based browser using the Chrome DevTools Protocol (CDP). This innovative tool allows users to give simple commands like “check my inbox” or “search for flights,” and the AI agent autonomously navigates the browser to fulfill these tasks without employing any conventional browser automation frameworks. It operates directly over a WebSocket connection to CDP, significantly reducing overhead by minimizing the amount of data processed with each command. Webact is particularly notable for its token efficiency, as it drastically reduces the volume of data exchanged during interactions, allowing agents to execute browser commands with minimal token usage—often by 80% or more compared to existing solutions like Playwright. This efficiency is achieved through its architecture, which emphasizes a perceive-act loop, where the agent plans, acts, and then perceives the results to make informed decisions. With a lightweight installation size of just 196 KB, Webact offers an agile alternative for developers who want seamless browser interaction without the baggage of larger libraries, making it a game-changer for AI applications in web browsing and automation.