AI Briefingjev-ultrafast: Complete Google Flights Booking in 7 Seconds: A Browser Agent Without Screenshots
browser-use/jev-ultrafast
About the project
Automate web browser manipulation by entering a single natural language goal. TypeSafe's Jev model simultaneously predicts the elements to click and the text to enter, eliminating the need to wait for multiple API calls as in traditional approaches. It completes a flight search from Zurich to London in just 7.1 seconds, including actual loading time.
The core innovation is reading the DOM structure directly instead of analyzing screenshots. All interactive controls on the page are converted into an indexed table and passed to the LLM, which selects targets specific to actions like CLICK or TYPE_TEXT. This reduces browser protocol calls from 1,092 to 101 and cuts total task time by 25%.
A small LLM intervenes only when text input is required to generate content. All other actions execute immediately based on structured state data, avoiding overhead from screen rendering or image processing. While complex UI elements such as Shadow DOM or canvas are not yet supported, it delivers sufficient performance for standard HTML form manipulation.
browser-use/jev-ultrafast
Fastest and cheapest web agent
Python
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.