GitHub · Browser agents
browser-use/jev-ultrafast
The fastest way to see the decide-don't-generate thesis in a real agent.
What it does
A browser agent with a dynamic, indexed action space: TypeSafe's Jev picks an operation and an element out of an index built from the live page, and a small LLM writes text only when the operation is TYPE_TEXT. The README's headline result is Zürich to London on Google Flights in 7.1 seconds, with the text generation and the page's loading waits included in that number.
The repository's own description: “Fastest and cheapest web agent”
How it works
The repository is a Python package (`jev_ultrafast/`) with `docs/`, `examples/`, `scripts/` and `tests/`, managed by uv. The action space is rebuilt per page rather than fixed, which is what lets Jev choose from an index of elements instead of emitting coordinates.
The repository, by the numbers
Read from the GitHub API on 2026-09-22. Stars and the last push move daily — quote them with the date, the way we do.
Numbers the author reports
These are the project's own figures, not our measurements — we keep ours on the statistics page and label them as ours. Read on 2026-09-23, with the sentence each figure appeared in:
- 7.1 sone flight search, start to finish
- “Zürich → London on Google Flights in 7.1 seconds. One natural-language goal, actual text generation, and loading waits included.” — repository README
What we checked
- the top-level tree: docs, examples, jev_ultrafast, scripts, tests
- pyproject.toml and uv.lock at the root — a uv-managed Python package
- LICENSE is MIT (Browser Use)
- the README states the 7.1-second Google Flights run and that a small LLM writes text only when the operation is TYPE_TEXT
What we did not check. We did not run the agent, so the 7.1 seconds and any cost figure are the authors'; the repository publishes no test results from that run.