# Jev Ultrafast

> A Python browser agent that picks indexed actions over the DOM

- Category: Agents & dev (`agents`)
- Tags: `free`, `open-source`
- Official: https://github.com/browser-use/jev-ultrafast
- Wunderlust: https://wunderlust.xyz/r/jev-ultrafast
- Image: https://wunderlust.xyz/media/jev-ultrafast-poster.jpg
- Video preview: https://wunderlust.xyz/media/jev-ultrafast.webm
- Added: 2026-11-05

## What it is

An open source web agent from browser-use that takes a natural-language goal and acts on a numbered table of page elements, without relying on screenshots.

## What it’s for

Use it to automate browser tasks with fewer calls and lower latency than image-based agents.

## Benefits

- Each step picks the operation and target in a single request
- A local inspector at 127.0.0.1:8766 shows elements, probabilities and executed actions
- The executor checks whether the page changed and whether the click is obstructed; model output never becomes a selector or JavaScript

## Caveats

- MIT license, but requires keys for the TypeSafe service and a text model (the example uses OpenRouter)
- According to the README, it is the fastest and cheapest web agent; the numbers are self-reported and not a general benchmark
- MVP version: no support for shadow roots, frames, canvas, file uploads or popup tabs

## Similar on Wunderlust

- [Ollama](https://wunderlust.xyz/r/ollama.md): Runtime for running language models on your own machine
- [Agent Browser](https://wunderlust.xyz/r/agent-browser.md): Tool that gives AI agents web browsing
- [Pizza Bot](https://wunderlust.xyz/r/pizza-bot.md): Local inbox for long-running AI agents, with an approvals queue
- [Playwright CLI](https://wunderlust.xyz/r/playwright-cli.md): Playwright's CLI, designed for agents to control the browser using few tokens
- [OpenDots](https://wunderlust.xyz/r/opendots.md): Open source template for always-on AI coworkers, each with its own computer
- [Unsloth Studio](https://wunderlust.xyz/r/unsloth-studio.md): App for running, chatting with and training AI models locally