Comal

Voice-first writing service. Hold down spacebar, talk, get a card with the transcript. Uses Sarvam for speech-to-text.

Run it

npm install
vim .env.local   # paste your Sarvam key
npm run dev

Without a key it falls back to mock mode so the UI is still testable.

How it works

• lib/recorder.ts — single-take MediaRecorder. Returns the full blob on stop.
• app/api/transcribe — server-side proxy to Sarvam (saaras:v3) so the key never reaches the browser.
• lib/session.ts — state machine (idle → listening → processing → post) and the TranscriptItem[] model. Each recording is its own card.

Built with: Next.js (App Router) + React 19.