Scribrrr

Real-time AI transcription for meetings.

Stack

• Frontend: Next.js, TypeScript, Tailwind, Zustand
• Backend: Fastify, TypeScript, Deepgram SDK, OpenAI SDK, Playwright
• Storage: In-memory (no database required)

Repo structure

apps/
  client/     → Next.js frontend (port 3000)
  server/     → Fastify backend (port 3001)

Quick start

1. Install dependencies

npm install

2. Set environment variables

Backend — copy and fill in apps/server/.env:

cp apps/server/.env.example apps/server/.env

PORT=3001
DEEPGRAM_API_KEY=your-deepgram-key
OPENAI_API_KEY=your-openai-key
FRONTEND_URL=http://localhost:3000

Frontend — copy and fill in apps/client/.env.local:

cp apps/client/.env.example apps/client/.env.local

NEXT_PUBLIC_BACKEND_URL=http://localhost:3001
NEXT_PUBLIC_BACKEND_WS_URL=ws://localhost:3001

3. Run the backend

npm run dev:server

4. Run the frontend

npm run dev:web

Open http://localhost:3000.

Testing the flow

1. Open the app → click New Session
2. Click Start Recording — transcript segments appear live as chat bubbles
3. Click Summarize last 2 min in the sidebar — calls OpenAI to summarize
4. Click Generate PDF in the sidebar — renders a report PDF via Playwright
5. Click Stop Recording when done

Deepgram configuration

Audio format settings are in apps/server/src/routes/stream.ts. Look for // TODO: comments. You may need to adjust:

• encoding (e.g. linear16, opus)
• sample_rate (e.g. 16000, 48000)
• channels

The frontend sends audio via MediaRecorder (webm/opus by default). If Deepgram needs raw PCM, switch to an AudioWorklet-based recorder on the frontend.

PDF generation

PDFs are saved to /tmp/generated/ and served by the backend at /files/. On first run you may need to install Playwright browsers:

cd apps/server
npx playwright install chromium

API routes

Method	Route	Description
GET	/health	Health check
POST	/sessions	Create a new session
GET	/sessions/:id	Get session details
GET	/sessions/:id/transcript	Get full transcript
POST	/sessions/:id/summaries/recent	Summarize last 2 minutes
POST	/sessions/:id/pdf	Generate PDF report
WS	/sessions/:id/stream	Live transcription WebSocket

What's not included yet

• Authentication
• Persistent database (everything is in-memory, resets on server restart)
• Speaker identification (just Speaker 0, Speaker 1, etc.)
• Production deployment config