Browser Use 1.0 ships persistent Python CDP runtime
A new engine, not an update
Browser Use published version 1.0.0 on November 10, 2026, at 02:04 UTC, according to the project's GitHub release page (GitHub). The release note is blunt about the scope of the change: "The agent now writes Python in one persistent REPL, straight on Chrome DevTools Protocol." That single sentence marks the end of the library's prior approach, which relied on a DOM serializer and a fixed action schema to translate agent intent into browser commands — both of which the 1.0 release removes (GitHub).
What changed under the hood
In the old architecture, an agent selected from a predefined menu of actions — click, type, scroll — each mapped through a serialized representation of the page's DOM. Version 1.0 discards that layer entirely. Instead, the agent writes Python code directly, and that code executes in a persistent REPL session wired straight into the Chrome DevTools Protocol (CDP), the same low-level interface Chrome exposes for browser automation and debugging tools. The practical effect is that the agent is no longer constrained to a fixed vocabulary of browser actions; it can express arbitrarily complex browser interactions as code, with the REPL retaining state across steps rather than resetting on each action.
Migration is not optional for 0.x users
Browser Use's own migration documentation does not describe 1.0 as an incremental update. It states plainly that "1.0 replaces the engine" and walks through a list of breaking changes for anyone running 0.x in production (Browser Use docs). Those changes include altered browser defaults, revised handling of API keys, and the removal of parameters that were previously injected automatically into agent calls. Teams with existing Browser Use deployments will need to audit configuration before upgrading, since code written against the 0.x action-schema model is unlikely to run unmodified against the new Python-REPL-on-CDP model.
Why it matters for agent builders
Browser Use is one of the more widely used open-source frameworks for giving AI agents control of a real Chrome browser, making this a developer-tools release rather than a consumer-facing launch. The shift to a persistent, code-writing runtime connected directly to CDP removes an abstraction layer that previously stood between agent reasoning and the browser itself — a design choice that collapses two prior stages (serialize DOM, map to action) into one (write and run Python). The release is attributed on GitHub to the account "gregpr07," the project maintainer credited with the publication (GitHub). No pricing, funding, or enterprise-adoption figures accompany the release; the available material documents the technical and migration changes only.


