The call is in another language. Your screen is not.

Know what to say while they’re still speaking — the answer is already on your screen, both sides translated under it. Works in any calling app.
Nobody on the call knows it’s there.

Try for Free

Five days free · no card, no account

A real call. The assistant’s answer, the transcript of both sides, the translation, and speech sent back into the call.
1
Install and open it

Windows 10 or 11, 64-bit. The installer isn't code-signed yet, so SmartScreen warns once — More info, then Run anyway.

2
Paste two keys

Deepgram for speech recognition, Groq for translation. Both register free, and the app links straight to each page.

3
Press Start session

That's it for on-screen work. VB-CABLE is only needed if you want the translated speech sent back into the call.

How it works

Everything you need while the call is running

The app runs at the Windows audio layer instead of inside one meeting service. That is why it hears the other person even when you're on headphones, and why it works with any calling app.

Suggests an answer from your own context

It waits for a finished question, then puts a short answer on screen while the pause still looks natural — and the full version while you're already speaking.

  • Uses your CV, a job description or a product sheet
  • One click researches the company: stages, reports, likely questions
  • Remembers what was said earlier in the call
AI Assistant
Short answer security interview
Map the roles and the expected access matrix first, then test horizontal and vertical privilege escalation on every endpoint.

Start from the tenant boundary: the same role in another tenant, then a lower role in your own. Check object ids in the path, in the body and in filters, and confirm the server decides, not the UI.

Hears both sides and translates on the fly

It captures your microphone and the system audio separately, turns each into text as the words come, and puts the translation on screen — or says it back into the call.

  • 27 languages, both directions
  • Original and translation side by side
Live · DE ↔ EN
Them
Wie würden Sie die Autorisierung in einer Multi-Tenant-API testen?
How would you test authorization in a multi-tenant API?
You
I’d start with a map of roles and tenant boundaries.
Ich würde mit einer Übersicht der Rollen und Tenant-Grenzen beginnen.

Answer in another language — in your own voice

Record 10 to 60 seconds once and the app builds a voice profile locally. Your translated speech then goes into the call sounding like you.

  • The sample and the profile never leave your computer
  • Local Piper voices and Microsoft Edge Neural voices as alternatives
Voice profile
Sample
00:42 recorded · stored locally · never uploaded
Outgoing speech
Ich würde mit einer Übersicht der Rollen beginnen.

Stays in front of you without getting in the way

A compact panel above the call window, adjustable in size and opacity, with a mode that hides the transcript and leaves only the answers.

  • Opacity from 20 to 100 percent
  • Panel stays on top of any window
Suggestion panel
Suggestion
Start with the map of roles and objects, then check access across tenants.
Opacity
Always on topAnswers only

Records the call and writes it down for you

Every session is recorded and transcribed as it happens, so you never take notes while thinking about your answer. It's all still there weeks later.

  • Each side recorded and transcribed on its own track
  • Transcript exported to a text file in one click
Call workspace
Session · 2 audio tracks
Transcript · Summary · Bookmarks · Recording

Answers what is on your screen, not only what was said

Press the middle mouse button anywhere in Windows and the assistant gets a snapshot as context — the diagram on the board, or the code they just opened.

  • Works from any app, and the shortcut is yours to pick
  • The shot goes to the AI provider on your own key and is not stored
Screen context
Captured
Middle click · the code review on screen
Middle click · the query on screen
Short answer
The token is compared with ==, so a wrong value still passes on a timing side channel.
The query builds SQL by concatenation, so the tenant filter can be dropped from the outside.
Privacy

Your calls never reach us

Nothing can sit on our side, because there is no side of ours. The only thing that leaves your machine is what you point your own provider keys at.

No account, no server of oursThe app runs on your computer. We never receive or store a second of your conversations.
Everything is stored locallyHistory, transcripts, recordings, settings, uploaded documents and your voice profile.
Your provider keys are protectedYour keys are stored encrypted and can only be read by your own Windows account.
Not fully offline — and we say so plainlyAudio goes to Deepgram for recognition; transcript, CV and job context go to the translation and AI providers you pick, all on your own keys.
Company research is a web searchWhen you press OSINT, only the vacancy text is sent to a web search service to collect public information about the company. Nothing from the call goes with it.
You control what runsRecording, translation, speech output and AI suggestions switch on and off separately. Deleting a call removes its transcript and recording.
Nothing here trains anythingYour calls, transcripts and documents are never collected, never pooled and never used to train a model. There is no pipeline for it, because there is no server of ours for them to reach.
No telemetry at allNo analytics, no crash reporting, no usage counters, no cookies. The app doesn't phone home — the only request it makes on its own is the licence check.
The panel is excluded from screen captureWindows itself is told not to hand the suggestion window to a recorder or a screen share, so it stays off the picture other people see.
The app checks itself before it startsEvery run verifies that the files have not been altered. A tampered copy refuses to start instead of running quietly.
The licence check knows almost nothingIt stores an anonymous identifier for the machine and the dates your licence starts and ends. No name, no email, no payment details, and nothing from your calls.
Open to inspectionEvery release ships a full list of the components it is built from, so a security team can review it without taking my word for anything.
Why this one

What you get that others do not

One app for every call you can't follow: work meetings, a language you are still learning, and the interview that has to go right the first time.

27languages, both directions
3modes: translator, companion, interview
0servers of ours in the path
Hiddenfrom screen sharing when you want it
10–60 sof speech to clone your voice
Anyone on daily cross-language callsClient calls, standups, sales, support — you follow all of it without asking people to repeat themselves.
People learning the language they work inYou get the translation, the gist, and the words and grammar behind what was just said.
Engineers interviewing abroadYou know the material. The language and the clock are what stand between you and the offer.

Not for faking expertise — read out an answer you don't understand and the first follow-up will expose it. This is a prompt, not a substitute for knowing your field.

Eleven AI providers, two speech engines and cloud voices — all on your own keys. One answers short, another answers in full.

FAQ

What people ask before installing

Will the other person hear or see that I am using it?

The app sends nothing into the call on its own. The only thing that goes into the call is what you turn on — the translated speech, if you enable it. The suggestion panel is an ordinary window on your computer, and you decide whether it appears when you share your screen. Test your own setup in advance, and if your company's rules forbid outside assistance, don't use it.

What else do I need besides the app?

Windows 10 or 11, 64-bit. Your own Deepgram key for speech recognition and Groq key for translation — registration is free, and the Ask me link in the header reaches me if you get stuck entering them. VB-CABLE is only needed if you want the translated speech sent back into the call; for on-screen translation and suggestions it isn't required.

Does it work with my calling app?

Yes, as long as the call runs through your computer. The app captures audio at the Windows device layer rather than plugging into one particular service, so it works the same with Zoom, Google Meet, Microsoft Teams, Discord and calls in a browser.

Where do my recordings and transcripts go?

They stay on your computer. There's no account, and no server of ours that your conversations pass through. That said, the product isn't fully offline: audio goes to speech recognition and text goes to the translation and AI services you connect with your own keys, on your own terms with those providers.

How fast does a suggestion appear?

Speech becomes text in about a second on a normal connection — measured across a real 45-minute interview, the median was 1.1 s and the fastest lines landed in 0.23 s. The short answer follows while the pause still looks natural; the detailed one arrives while you're already speaking. Exact numbers depend on your connection, the recognition service and the model you choose.

Which AI providers can I use?

Eleven of them, and you pick one for the short answer and another for the detailed one. Through a local CLI you already own: ChatGPT / Codex, Claude Pro, Grok CLI. With your own API key: Google Gemini, Groq, Cerebras, NVIDIA NIM, Cloudflare Workers AI, OpenRouter, Token Harbor, and Antigravity through a local proxy. Auto races several of them and takes whichever answers first, falling back when one is rate-limited or down. Speech recognition is Deepgram Nova-3 or Google Gemini Transcribe Live; translation runs on any OpenAI-compatible endpoint or on Google Gemini.

Can it see what is on my screen?

Only when you ask it to. Press the middle mouse button anywhere in Windows and a single snapshot of the screen is sent to your AI provider along with the question, so the answer is about the diagram, task or code in front of you. The trigger is configurable — the middle button by default, or a side button, or any key with modifiers. Nothing is captured while you're not pressing it, the image isn't saved to disk, and it goes to the same provider on the same key as every other answer.

Can it help in a job interview?

Yes, that's one of the three modes. A translator explains what was said and the words behind it, a companion just follows the conversation, and an interview assistant reads your CV and the job description and drafts a short answer plus a detailed one. Beyond that it records any meeting and writes it down, shows a teleprompter with a script you prepared, and can answer in your own cloned voice.

What happens if my connection drops mid-call?

The licence keeps working for a full day offline, so a flaky hotel network or a dropped VPN doesn't throw you onto an activation screen in the middle of a call.

Can I install it on a second computer?

You get one five-day demo per computer, bound to that machine. A second computer gets its own five days from the same installer. Reinstalling Windows on the same machine doesn't hand you a fresh period. When the five days end, or if you need it on more than one machine, use the Ask me link in the header and I will issue a full licence.

Windows warns me about the installer. Is that normal?

Yes. The installer isn't code-signed yet, so SmartScreen warns the first time — press More info, then Run anyway. After the first launch the warning doesn't come back.