Custom Cloud Models

If you use a transcription service that Echo does not include by default, you can add it as a custom cloud model. Any service that provides an OpenAI-compatible transcription API will work.

Adding a Custom Model

1. Open Echo and go to Settings (click the gear icon in the sidebar, or press Cmd + ,).
2. Select the Transcription tab.
3. In the model list, click the Custom filter at the top.
4. Click Add Model.
5. Fill in the form fields:
   • Display Name -- a label you will see when selecting this model (for example, "My Whisper Server").
   • API Endpoint -- the full URL of the transcription endpoint (for example, https://api.example.com/v1/audio/transcriptions).
   • API Key -- your authentication key for the service.
   • Model Name -- the model identifier expected by the API (for example, whisper-1 or large-v3-turbo).
   • Multilingual Model -- turn this on if the model supports multiple languages, or off if it is English-only.
6. Click Add Model.

Your new model will appear in the Custom section of the model list and can be selected like any built-in model.

Editing a Custom Model

1. Go to Settings > Transcription and select the Custom filter.
2. Find the model you want to change and click its Edit button.
3. The form will open with the current values pre-filled. Update any fields you need to change.
4. Click Update Model to save your changes.

Removing a Custom Model

1. Go to Settings > Transcription and select the Custom filter.
2. Find the model you want to remove and click its Delete button.
3. Confirm the deletion when prompted.

If the deleted model was your active transcription model, you will need to select a different model before your next recording.

Requirements

• The API must be OpenAI-compatible -- it should accept audio files at the endpoint you provide and return transcription text in the same format as OpenAI's transcription API.
• You need a valid API key for authentication.
• The endpoint must be reachable from your Mac (check any firewall or VPN settings if you are connecting to a self-hosted server).

Tips

• Custom models work everywhere in Echo, including in Adaptive Awareness profiles. You can assign a custom model to a specific profile just like any built-in model.
• If you run a local transcription server (for example, a self-hosted Whisper API), use http://localhost or your local IP as the API endpoint.
• Double-check the model name your provider expects. Getting this wrong is the most common reason a custom model fails to return results.

Related

• Cloud Transcription Services -- built-in cloud providers included with Echo
• Local Models -- offline transcription with downloaded models
• Adaptive Awareness Profiles -- assign models per profile