AI Agent News/stories/browser-use-1-0-0-replaces-its-engine-with-a-python-repl-agent.md

Browsers

browser-use 1.0.0 replaces its engine with a Python REPL agent

browser-use 1.0.0 is out, and its agent now writes Python in a persistent REPL instead of choosing one DOM action per step. The library shrinks from 68,815 to 13,738 lines of Python, and the main 0.x classes and methods still work.

Source: github.com/browser-use/browser-use/releases/tag/1.0.0

The release swaps out the engine underneath the library. The agent writes Python in one persistent REPL, using browser-harness helpers and raw CDP, and runs on Pydantic AI. The default model is bu-fast on the Browser Use gateway, and Browser() now opens a cloud browser, with Browser('chromium') for a local one.

Several defaults changed. run(max_steps=...) now defaults to 100 instead of 500, and a run has a 1700 s timeout. Once configuration checks pass, run() returns status='error' with result.error instead of raising, and any removed 0.x import raises an ImportError that points to MIGRATING.md.

In a BU Bench v2 run of 60 tasks with three repetitions on cloud browsers, 1.0 scored 62.5 against 56.3 for Browser Use Pi in the same window, though the source says the gap is not significant once provider errors are excluded. The benchmark ran on the 1.0.0rc2 build. Known issues include exact dependency pins that can conflict with other projects and no viewport emulation on cloud browsers.

Why it matters if you build agents

Pin the exact version with == and read MIGRATING.md before upgrading, since 0.x code that relied on localhost or private IPs needs Browser('chromium') now. Check result.error after run(), because failures no longer raise exceptions.

If your project pins pydantic, mcp or google-genai to other versions, 1.0 may not install beside them, so test the dependency set before upgrading.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe