Show HN: Browser Tools SDK – an optimal browser harness for agents (libretto.sh)

A new Browser Tools SDK has been launched, enabling AI agents to interact seamlessly with web pages through a real browser using Playwright. This SDK provides six essential tools that allow agents to open a browser, read a webpage’s content, and execute necessary commands while minimizing overhead costs. It supports various platforms, including Kernel and Browserbase, and offers compatibility with existing AI SDKs, with future support for Pi on the horizon. The SDK's architecture enables AI agents to generate text responses—like finding top stories from Hacker News—by leveraging a compact page snapshot and efficiently running specific Playwright code. This development is significant for the AI/ML community as it streamlines the process of web interaction for AI agents, facilitating tasks that require navigating complex web structures without the need for raw HTML parsing. The technology enhances an agent's ability to work with live pages through a new compact accessibility tree and dynamic snapshot diffs, simplifying the workflow while preserving crucial technical capabilities. The Browser Tools SDK is available under the MIT license, allowing for easy installation and integration into existing frameworks, thus broadening the potential for AI applications in real-time web environments.