Jev Voice Browser
A browser that starts resolving voice commands while the user is still speaking.
What Jev decides
What the user intends, which page target it applies to, whether the command is complete, and whether it is destructive.
Each partial transcript is sent to Jev, which classifies the intent and selects targets from the current page while separate checks decide whether the command is complete, executable, or destructive.
- Jev role
- Resolve partial speech into a bounded browser intent, target, and safety decision.
- Input
- Partial transcript, current URL, tabs, visible interactive elements, and recent browser state.
- Output
- Choice distributions for intent and targets, Noul probabilities for command state and risk, and a Score for scroll amount.
- Decision frequency
- Once for each incoming partial speech transcript.
- Fallback
- Wait for more speech or ask for confirmation; deterministic Playwright code performs the selected action.
How it works
Reported performance
Reported Jev latency
250–350 ms
Resolve one partial voice transcript against the current page.
Author reported · Source reviewedCaptured Sep 21, 2026
Reported cost per request
0.0002 USD
One partial-transcript decision request.
Author reported · Source reviewedCaptured Sep 21, 2026
Overview
Jev Voice Browser controls a headed Chromium session through Playwright. It extracts a compact view of visible interactive elements and sends every partial speech transcript to Jev, allowing the browser to prepare or execute safe actions before the final transcript arrives.
Intent, site, element, and tab are bounded Choice outputs; completeness, command validity, and destructive risk are Noul decisions; scrolling uses Score. Destructive actions require confirmation. The prototype runs on localhost without authentication and should not be exposed as a public service.
- Language
- JavaScript
- License
- MIT
- Author
- Moritz Kremb
Sources
Related projects
jev-ultrafast
A browser agent where Jev selects an operation and a compatible page element from a live indexed action space.
What Jev decides
Which supported browser operation to run, and which compatible visible element to target
TypeSafe Mario
Jev plays Super Mario Bros. from structured emulator telemetry rather than screenshots.
What Jev decides
The next controller action, whether jumping is useful now, and the immediate danger level.
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.