Prerequisites
Both surfaces share the same dictation pipeline (Deepgram Nova-3 transcription + Claude post-processing) and the same bring-your-own-key model. Set these up once, then continue to whichever surface you chose.
| Requirement | Purpose | Shared? |
|---|---|---|
| ffmpeg | Microphone recording | VS Code extension only — the macOS app captures audio natively via cpal and does not need ffmpeg |
| Deepgram API Key | Nova-3 transcription | Shared by both surfaces |
| Anthropic API Key | Claude post-processing | Shared by both surfaces |
Installing ffmpeg
Required for the VS Code extension. Skip this if you're only installing the macOS app.
Download from ffmpeg.org and add to PATH, or via Chocolatey:
API Keys
On first use, Verba prompts for your API keys. On VS Code they're stored in SecretStorage; on macOS they're stored in the system Keychain — never in plaintext, and only sent to the providers you use: Deepgram (transcription), Anthropic (cleanup), and — in VS Code only — OpenAI if you enable embeddings for context-aware templates.
- Deepgram API Key — Get one at console.deepgram.com
- Anthropic API Key — Get one at console.anthropic.com
Offline Alternative
Don't want to send audio to Deepgram at all? The VS Code extension supports fully offline transcription via whisper.cpp, running locally with no network calls. See Offline Transcription. This is currently VS Code–only.