Desktop app
The desktop binary ships with its own UI (no internet connection required to open it after the first sign-in) and talks to the web API over HTTPS with a short-lived JWT.
Audio capture
- Microphone (interviewer): captured from the system microphone. The app asks for permission the first time you start a recording.
- System audio (candidate): captured via ScreenCaptureKit on macOS and WASAPI loopback on Windows. macOS asks for Screen Recording permission the first time — grant it once in System Settings → Privacy & Security → Screen Recording.
- Linux: microphone only. System-audio capture isn't supported because there's no portable equivalent of ScreenCaptureKit / WASAPI loopback.
What leaves your machine
Audio frames are streamed to our STT proxy (a Cloudflare Worker) over
WebSocket; the proxy forwards them to the speech-to-text provider (Deepgram
by default, AssemblyAI / RevAI as fallback). Final transcript lines come
back the same path and are POSTed to /api/edge/analyze-chunk for AI
analysis. Nothing else leaves the machine.
Where things are stored
- Access token (Clerk JWT): encrypted at rest by the OS keychain via
tauri-plugin-store. Cleared on sign-out. - Transcripts and scorecards: in the cloud database, scoped to your organization. The desktop app does not keep a local copy by default.
- Audio: never persisted to disk (live mix → WebSocket → STT, then discarded).
Platform notes
macOS
- Requires macOS 14 Sonoma or newer.
- First launch: Right-click → Open (one-time Gatekeeper prompt) — even though the binary is signed and notarized, the very first launch from outside the App Store benefits from explicit consent.
- The Tauri auto-updater verifies releases with a minisign signature; the public key is baked into the binary.
Windows
- Requires Windows 10 or 11, 64-bit.
- SmartScreen may warn on first launch (we are working on Authenticode signing). Click More info → Run anyway.