Using Local Tiny Whisper for Mixed Web Mode

OpenScribe mixed web mode supports fast local transcription with Whisper and separate larger-model note generation.

1. Configure environment

Edit apps/web/.env.local:

TRANSCRIPTION_PROVIDER=whisper_local
WHISPER_LOCAL_URL=http://127.0.0.1:8002/v1/audio/transcriptions
WHISPER_LOCAL_MODEL=tiny.en
WHISPER_LOCAL_BACKEND=cpp
WHISPER_LOCAL_TIMEOUT_MS=15000
WHISPER_LOCAL_MAX_RETRIES=2

2. Start everything (one command)

pnpm dev:local

This command will:

• create/check backend venv
• install backend deps if missing
• create apps/web/.env.local if missing
• start local Whisper (tiny.en + whisper.cpp, using acceleration when available)
• start Next.js web app

3. Verify health

curl http://127.0.0.1:8002/health

You should see JSON including the active backend and model.

Notes

• Tiny English model maximizes speed; use WHISPER_LOCAL_MODEL=base if you need more accuracy.
• Keep note-generation model settings independent from transcription settings.