Yogurt Worker

This repository contains the Cloudflare Worker implementation for Yogurt, a macOS notes app that enhances meeting notes by capturing and transcribing both system audio and microphone input.

The Worker handles audio transcription and note enhancement using Cloudflare Workers AI, providing two key endpoints that the Yogurt macOS app relies on.

Features

• Real-time audio transcription using whisper-large-v3-turbo
• Context-aware note enhancement using llama-3.3-70b-instruct-fp8-fast
• Multi-stage processing pipeline:
  • Transcript to comprehensive notes conversion
  • Points of emphasis identification
  • Action item extraction
  • Final enhanced notes generation
• Zero data retention - processes data in-memory only
• Optional Cloudflare Access authentication support

Requirements

• A Cloudflare account with Workers AI enabled
• Wrangler CLI (for development)

API Endpoints

POST /transcribe

Handles audio transcription requests.

Input:

• Raw audio data in WAV format
• Automatically filters out background noise using VAD (Voice Activity Detection)

Output:

{
  "text": "Transcribed text content"
}

POST /transcription-notes

Converts raw transcript text into structured, comprehensive notes.

Input:

{
  "transcript": "Raw transcript text"
}

Output:

• Server-sent events (SSE) stream of formatted notes

POST /points-of-emphasis

Identifies important points that appear in both user notes and transcript.

Input:

{
  "userNotes": "User's markdown notes",
  "transcriptNotes": "Processed transcript notes"
}

Output:

• Server-sent events (SSE) stream of emphasized points

POST /action-items

Extracts action items from meeting notes and transcript.

Input:

{
  "userNotes": "User's markdown notes",
  "transcriptNotes": "Processed transcript notes"
}

Output:

• Server-sent events (SSE) stream of action items

POST /enhance (Deprecated - use the endpoints above)

Processes and enhances meeting notes using transcript context.

Input:

{
  "notes": "User's markdown notes",
  "transcript": "Meeting transcript"
}

Output:

• Server-sent events (SSE) stream
• Enhances notes using prompts

Deployment

1. Clone the repository:

   git clone https://github.com/jonesphillip/yogurt-worker.git
   cd yogurt-worker

2. Install dependencies:

   npm install

3. Deploy to Cloudflare Workers:

   npx wrangler deploy

Development

For local development:

npx wrangler dev

Integration with Yogurt

To connect this worker with the Yogurt macOS app:

1. Deploy the worker to your Cloudflare account
2. Copy your worker's URL (e.g., https://your-worker.username.workers.dev)
3. Open Yogurt settings and paste the URL
4. (Optional) Add your Cloudflare Access service token if configured

Development

For local development:

npx wrangler dev

This will start a local development server that you can use for testing.

Security considerations

• Consider enabling Cloudflare Access for additional security
• The worker processes all data in-memory and does not persist any information
• Audio data is processed in chunks and immediately discarded

Related projects

• Yogurt - The main macOS application