Get started
Menu
Browse documentation

Start here

Connect

Use Yep Anywhere

Get help

Yep Anywhere docs

Notifications and voice

Configure browser push notifications and choose browser, hosted, or local speech-to-text input.

Notifications keep a waiting agent from sitting idle. Voice input makes short responses and longer instructions practical from a phone.

Browser push notifications

Open Settings → Notifications, grant browser permission, and subscribe the current browser profile. Yep Anywhere generates and stores its own VAPID keys on the server.

The server decides which events deserve notification; the receiving browser decides whether to show them based on focus and the visible session. A connected tab is not automatically treated as visible.

If notifications do not arrive:

  1. Check the browser’s site permission and operating-system notification settings.
  2. Confirm the current browser profile is subscribed.
  3. Send a test notification from Settings.
  4. Check that the service worker is active and the server has VAPID keys.

Native Android push is in development, with iOS planned afterward. Browser Web Push remains the published phone-notification path; neither native mobile app is published.

Browser voice input

Browser-native recognition requires no server key when the browser supports it. Open the microphone menu, choose the browser backend, and grant microphone permission.

Hosted speech backends

Server-mediated Deepgram and xAI transcription use dedicated server environment keys:

export YEP_STT_DEEPGRAM_API_KEY="deepgram-..."
export YEP_STT_XAI_API_KEY="xai-..."

Speech keys are consumed by Yep Anywhere and stripped from child agent environments. A dedicated STT key is preferred over a general provider key so speech billing and agent billing stay separate.

Local speech backends

Local Whisper, Parakeet, and NeMo are opt-in:

export YEP_VOICE_BACKENDS=ya-whisper,ya-parakeet

The first startup prepares the selected local environment and downloads its model. NeMo is the heavier optional backend. A backend that fails setup or validation is hidden and the server logs a repair hint rather than offering a known-broken microphone choice.

Disable the entire voice surface with VOICE_INPUT=false when microphone input is not appropriate for an installation.