Your own words do not trigger answers
Only the interviewer's turns are considered as questions. Without separation, saying “so how did we handle that” out loud would generate an answer to yourself mid-sentence.
Live transcription
The call becomes text. The audio does not become a file.
Everything else depends on this step: there is no question to detect and no answer to write until the call is words. Here is how it is heard, who is told apart from whom, and where the audio goes — which is nowhere.
Free, however long the call runs.
Two places it can happen
On our server, or on your own PC
The same transcript either way. The difference is where the audio exists while it is being read, which is the part worth caring about.
| Aspect | Browser | Windows app |
|---|---|---|
| How it hears the call | You share the call's tab or window, with its audio | Directly, from whatever plays the call |
| Where it is transcribed | On our server | On your PC |
| Does the audio leave your machine | Yes, in short pieces, and is discarded after being read | No. Never. |
| Is a recording kept | No | No |
| Setup each time | Pick the tab to share | None |
| Works on | Chrome or Edge on a computer, best | Windows 10 or 11, 64-bit |
Get the Windows app → It signs in with the same account and runs the same sessions; you can move between the two mid-call.
Who said what
The interviewer and you are kept apart
Not a nicety. Three things depend on it.
Context stays straight
The last few turns are given to the model so a follow-up makes sense. Which of those turns were yours changes the meaning of “and how would you test it?” entirely.
The transcript is readable afterwards
A conversation you can follow, with each suggested answer sitting against the question that prompted it. A single undifferentiated block is not worth keeping.
The audio
What happens to it, precisely
Worth being exact about, because this is the part people are right to be careful with.
In the browser
Short pieces of audio are sent to be transcribed, read, and discarded. No file is written, nothing is retained for training, and there is no recording of your interview in existence afterwards.
In the Windows app
The audio is transcribed on your PC and never leaves it. Only text goes to the server, and only when an answer is being written.
Screenshots
When you send your screen, the image is read and not kept. The question text and the answer are part of the session; the picture is not.
The transcript itself
Text, in your account, under your control: kept indefinitely, kept thirty days, or deleted the moment the notes are written. The full detail →
When it mishears
Accuracy matters less here than you would expect
A transcriber will get a surname, an accent or an unusual library wrong. Three reasons that is survivable.
The answer uses the sense, not the spelling
A question with one word mangled still asks the same thing, and the answer is written from the question as a whole plus the surrounding conversation.
You can take over
Type or paste the question yourself at any point. Faster than repeating yourself to a laptop, and exact.
You can correct the answer
Shorter, more detail, more technical: one press, and it rewrites with what you have already said taken into account.
Questions
About transcription
How does it hear the interviewer in a browser?
You share the call's tab or window with its audio, the way you would share a screen with anyone. The browser then hands the assistant the same audio the call is playing. Chrome and Edge on a computer handle this best; it is a browser capability, not something the product can work around.
And in the Windows app?
Directly, from whatever is playing the call — Teams, Zoom, Meet, a browser tab, a phone on speaker. Nothing to share and nothing to remember to set up, which is the main practical reason people move to the app.
Is my call audio stored?
No. Audio is turned into text and discarded; there is no recording to keep, download or leak. In the Windows app it never leaves your PC at all, because the transcription happens there.
Does it know who is speaking?
Yes — the transcript separates the interviewer from you, which matters for two reasons. Only the interviewer's turns are considered as questions, so your own talking does not trigger answers. And afterwards the transcript reads as a conversation rather than a wall of text.
What does transcription cost?
Nothing, however long the call runs. Listening is free and you are charged for answers. A per-minute charge would mean watching a clock during an interview, which is the last thing anyone needs to be doing.
Do I need to tell the interviewer?
Possibly, and you should find out. Recording or transcribing a call can require the other side's consent depending on where you both are, and some employers forbid assistance of any kind. That judgement is yours and the responsibility comes with it.
What if the transcription gets a word wrong?
Technical vocabulary and accents both cost accuracy, and no transcriber is exact. It matters less than you would think, because the answer is written from the sense of the question rather than from every word of it — and where it does matter you can type the question instead, or refine the answer in a press.
Read next
Related pages
Try it on a call with a friend.
Share the tab, talk for two minutes, and read the transcript back.
Create your account