How sessions work:
The candidate runs two calls simultaneously: their real interview + a support call with you. They share their screen with audio so you hear the interviewer. You share your screen back so they see AI answers. No installs needed on their end.

Audio setup: Configure via 🎚️ Device — use BlackHole (macOS) or VB-Cable (Windows) to route the interview call's speaker output into this app. Set the interview app's speaker to BlackHole/VB-Cable so only interviewer audio goes through — not all system sounds.

Prep: Load resume + JD via 🎯 Prep for context-aware answers. Set interview type via the badge above or let it auto-detect.

Speed: Deepgram ~300ms transcription · GPT-4o-mini + Claude + Gemini in parallel = 1-2s answers.

Hold AI: Press H when candidate is speaking to pause AI without stopping capture. In loopback mode this is usually not needed.

Define who is in the room. Assign roles so the AI knows who's asking questions vs answering. Interviewees are transcribed but never trigger AI responses.

Pre-load topics to give AI better context. Topics are included in every AI query.

The candidate maintains two calls at once: their real interview call, and a support call with you. They share their screen (+ audio) in the support call so you can hear the interviewer. Answers appear on your screen, which the candidate reads via your screen share back to them.

Load the resume and job description so AI can give specific, grounded answers instead of generic ones.