Jev Ultrafast
Browser Use's browser agent where Jev picks an operation and page element in one request and a small LLM only writes text to type.
Overview
Jev Ultrafast is a browser agent from Browser Use with a dynamic, indexed action space. Each observation produces a fresh table of interactive elements, and the operations are CLICK, TYPE_TEXT, SELECT, SCROLL_UP, SCROLL_DOWN, WAIT, DONE and BLOCKED. One TypeSafe request asks for the operation plus speculative target questions (click target, type target and select target when present), each containing only compatible elements, so two decisions take one network round trip and only the matching target executes. A small LLM is called only when the operation is TYPE_TEXT.
There are no site-specific scripts or prepared field strings in the policy. The headline example completes a Zurich to London search on Google Flights from one natural-language goal in about 7.1 seconds, including text generation and page loads, with the outcome verified independently. The repository includes the library, a local inspector, a demo video and performance measurements. It has spawned several forks and wrappers, including MCP servers and UI-testing frameworks. Browser Use also advertises a cloud waitlist.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which operation should the agent perform next? | CLICK · TYPE_TEXT · SELECT · SCROLL_UP · SCROLL_DOWN · WAIT · DONE · BLOCKED |
| Choice | Which compatible element is the target for this operation? | — |
How it works
- Observe the page and build an indexed element table
- One Jev request returns operation and speculative targets
- Execute the matching target; call a small LLM only for TYPE_TEXT
- Repeat until DONE or BLOCKED
Strengths
- Two decisions per network round trip
- Cannot target elements that are not on the page
- Authors report a Google Flights task in about 7.1 seconds
- No site-specific scripts
Caveats
- Requires a TypeSafe key plus a small LLM for typing
- Timing measured on the authors' examples
Getting started
git clone https://github.com/browser-use/jev-ultrafast.git && cd jev-ultrafast && uv sync, then copy .env.example and add keys.At a glance
- Type
- Library
- Status
- beta
- Language
- Python
- Runtime
- Library / browser agent
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Browser automation: browser-use's ultrafast agent where Jev decides each next action and element to click, calling a language model only when text must be typed.
yibie/awesome-jevBrowser agent from [Browser Use](https://github.com/browser-use). Jev picks an operation and a DOM element in one request; a small LLM writes text only for `TYPE_TEXT`. Zürich → London on Google Flights in ~7s. Library, local inspector, and measurements included.
AnotiaWang/awesome-decision-modelsA browser Agent that uses Jev to choose actions and page controls, calling a text model only when input is needed.
logicrw/awesome-jev-projectsFastest and cheapest web agent.
wh000wh000/awesome-jev-liveBrowser agent where Jev selects an operation and compatible DOM target, and a separate LLM supplies typed text.
cobanov/awesome-jevBrowser Use's ultrafast agent: Jev picks the operation and DOM element in one request; a small LLM only writes text when typing is needed.
hellogumbo/awesome-jevFast browser agent from Browser Use. Jev decides each step and which element to act on; a small model is called only when text needs to be typed. The authors report a full Google Flights search in about 7.1 seconds.
MrJev/awesome-jev
Frequently asked questions
When is a language model used?
Only when the chosen operation is TYPE_TEXT, to write the text.
How fast is it?
The authors report a full Google Flights search in about 7.1 seconds.
Can it click something that is not on the page?
No. Targets are chosen only from the observed element table.
Related projects
Lets coding agents run whole goals in your signed-in Chrome on macOS in background tabs, asking Jev or Claude Haiku only for real choices.
ProjectbetaComputer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.
ProjectbetaPythonLocal stdio MCP server that executes whole browser and macOS goals with a Jev-driven observe-decide-act loop.
MCP serverbetaTypeScriptPython decision layer for browser and desktop automation: code observes, acts and verifies; the model makes one typed choice per step.
ProjectbetaPythonBrowser and computer use for coding agents via CLI and MCP, with Jev's calibrated judgments and a code-owned control loop.
MCP serverbetaTypeScriptCodex plugin adding a bounded Jev decision loop to Computer Use for browser tabs and native macOS apps.
Projectbeta