Your Server, Your Model
Most VocaMac speech engines stay on this Mac. Custom Endpoint is the opt-in exception: when you select it, each finished recording is uploaded as a WAV file to a Whisper-compatible HTTP endpoint you configure, and VocaMac pastes the transcript the server returns.
Use it when you already run whisper.cpp, an OpenAI-compatible audio API, or another self-hosted speech stack and want VocaMac’s hotkeys and paste flow without downloading another local model.
What It Speaks
Open Settings → Speech Model, select Custom Endpoint, and fill in:
| Setting | Meaning |
|---|---|
| Endpoint kind | OpenAI-compatible (POST <base>/v1/audio/transcriptions) or whisper.cpp server (POST <base>/inference) |
| Base URL | Your server’s origin. HTTPS for anything outside this Mac or your local network. A trailing /v1 on OpenAI-style bases is fine — VocaMac normalizes it. |
| Model | Sent as the OpenAI model field when that kind is selected; whisper.cpp ignores it. |
| API key | Optional Bearer token, stored in Keychain. Never put credentials in the URL. |
Recordings leave this Mac only while Custom Endpoint is the selected model. Switching back to any local engine returns to on-device transcription.
Honest Limits
- Vocabulary hints stay on this Mac; they are not uploaded with the recording.
- Redirects that would downgrade HTTPS to cleartext HTTP are rejected, even to localhost.
- There is no Voca-hosted speech cloud — the destination is whichever server you typed in.
Private by Design (for local engines)
Custom Endpoint is opt-in and explicit. Local engines keep audio on your Mac after their models are available. If you do not select Custom Endpoint, nothing in this feature applies.
Ready to try Custom Endpoint?
Install VocaMac