Sokrates listens to both sides of the call, transcribes live, and streams an answer you can glance at or read aloud — in an overlay that never appears in a shared screen.
Mac with Apple Silicon, or Windows 10/11
The browser version, on a phone, tablet or computer
| Question | Desktop app | Browser version |
|---|---|---|
| Where it runs | Mac (Apple Silicon) or Windows 10/11 | Any modern browser: Safari, Chrome, Edge |
| How it hears the interviewer | Straight from the computer’s sound; headphones are fine | Through the device’s microphone; the call has to be on speakers |
| In a screen share | Hidden: excluded from screen capture | Visible, like any browser tab |
| A written question | One shortcut takes a screenshot | On a phone, a photo of the screen with the camera |
| Together | A phone with the browser version can mirror the desktop app’s answers under your webcam and press its buttons | |
| Your plan | One account and one plan for both, Free or Pro | |
Free plan: 10 minutes a day, no card · Apple Silicon · Windows 10/11 · browser
I’d keep every positive and downsample the negatives, then use a class-weighted loss so the model still sees the true prior. I’d evaluate on PR-AUC rather than accuracy, and pick the threshold on the precision–recall curve for the recall we actually need at serving time.
Sokrates captures system audio, so anything that plays sound on your computer works. No plugin, no bot joining the call.
A short, fixed answer shape you can use mid-sentence. Everything else is there when the follow-up comes.
Your mic and the system audio are transcribed as two channels, labelled Interviewer and You. Answers come back in the interview's language: 45+ languages, right-to-left for Hebrew and Arabic. All languages →
Every answer opens with 3–5 bullets to glance at, then 50–90 words of first-person prose to read aloud. Coding gets Say · Code · Explain; design gets Design · Say · Trade-offs.
One click adds 250–350 words that build on the answer instead of restating it. In a system-design round it walks the next stage the interviewer will ask about.
Cmd/Ctrl+Shift+S captures the editor or whiteboard. A coding question comes back as runnable code with a two-bullet explanation and the complexity.
Résumé, job description, and the documents you upload. Company and role are set per session. Answers draw on your documents and are told to stick to them. AI can still get details wrong, so check what it says about your own work before you use it.
The overlay is excluded from screen capture at the operating-system level, so it never appears in a shared screen or a recording, on macOS and on Windows. How it works →
Sessions are saved to your local history. Each ends with a recap, a score out of 10, and what to tighten next time.
The browser version runs a session on any phone, tablet or computer with nothing to install, listening through the device's microphone. Unlike the desktop app it shows in a screen share. On a phone it can also mirror the desktop app under your webcam. The browser version →
Set up once before the call. During the call, three shortcuts do everything.
macOS (Apple Silicon) or Windows 10/11. One account works on both.
Google or email. The free plan starts immediately, no card.
Résumé, job description, notes. Set the company, role, and round type for this session.
Turn the mic on. Ask, take a screenshot, or let it answer as questions land. Go Deeper when they follow up.
Live answers as the interviewer talks. Scan bullets to glance at, Say prose to read aloud.
Learn more →An AI interviewer asks one question at a time and scores each answer out of 10.
Learn more →Key points for the moment during a work call, then a recap with action items.
Learn more →Runs in any browser, nothing to install. It shows in a screen share, unlike the desktop app; on a phone it can also mirror the desktop app under your webcam.
Learn more →Free is enough to take it into a real call. Pro is enough to run the whole interview, and the next one.
10 minutes of live transcription and 3 screenshots a day, every mode, answered by Claude Haiku 5.5, in the desktop app or the browser. No card, and it resets every day.
Download freeUse it for every interview, for as long as your search runs.
Renews every 3 months until you cancel. You see the total before you pay.
Hours of listening, paid once.
One payment. Nothing renews.
Prices in USD.
Free to download, no card. One live session a day on the free plan; Pro when you want the caps gone.