Video unavailable. Open the source below.
NewJev Voice Browser
A voice-browser demo combines speech input and Jev-selected actions in a rapid interaction loop.
Added to Jevfast
How it uses Jev
Each speech partial is paired with a page snapshot and recent action context. A single request asks typed questions about intent, target, site, command completeness, whether speech is directed to the browser, destructive risk, and other conditional fields. Code applies thresholds and executes or waits; candidate spans supply exact text and URLs.
What Jev decides
Illustrative Jev decisionExample answers · not a recorded Jev response · Source ↗
Question 1 · complete
YOUR APP
INSTRUCTION
Has the user finished an actionable spoken command?
STATE
Partial transcript, current page and elements, recent actions, and extracted text or URL candidate spans.
JEV · NOUL
YesNo
Question 2 · destructive
YOUR APP
INSTRUCTION
Would carrying out the command have a hard-to-undo side effect?
STATE
Partial transcript, current page and elements, recent actions, and extracted text or URL candidate spans.
JEV · NOUL
YesNo
Question 3 · scroll amount
YOUR APP
INSTRUCTION
Score requested scroll distance from a little to the end of the page.
STATE
Partial transcript, current page and elements, recent actions, and extracted text or URL candidate spans.
JEV · SCORE
0
App workflow
- Stream speech
Receive partial transcripts from the browser microphone.
- Snapshot context
Attach page elements and recent actions to the utterance.
- Ask Jev
Batch intent, target, site, completion, command, and risk questions.
- Apply policy
Use thresholds to act, wait for more words, ask, or ignore.
- Show decision
Update the controlled browser and display decision evidence in the control page.
Why it is interesting
Speech can be acted on before the sentence ends, with typed judgments controlling safety and candidate selection while text remains verbatim from speech spans.