AI engines overview
Transcription, translation, summaries, and the voice can run on your device or in the cloud; how to choose, and how the app switches automatically when something fails.
- macOS
- iPhone
- iPad
Each capability in Sonasi Meetings uses its own engine, and you choose each one:
| Capability | Default (Apple, on-device) | Models installed on your Mac | Cloud (your API key) |
|---|---|---|---|
| Transcription | Yes | No | OpenAI, ElevenLabs |
| Translation | Apple Translation | Yes (Mac only) | OpenAI, Anthropic*, OpenRouter, Ollama, custom |
| Summary | Apple Intelligence (macOS / iOS 26+) | Yes (Mac only) | Same as Translation |
| Voice (Mac only) | Apple, Sonasi TTS | Sonasi TTS | ElevenLabs, OpenAI, VieNeu |
* Anthropic is available only on iPhone and iPad.
What to choose
- Start with Default. It runs on your device, is free and private, and needs no setup.
- Switch to Cloud when you need higher accuracy: regional accents, several people talking over each other, noisy rooms, or Vietnamese (Apple's recognizer handles Vietnamese at lower quality).
- Use a Local model (Mac) if you want to translate or summarize with a model stronger than Apple's without sending data out. See Models on your Mac.
The On device or Sends to [provider] label always appears next to each engine so you know whether data leaves your device.
How to choose an engine
Mac
Go to Settings → Intelligence & engine. There are two quick modes:
- Standard: everything runs on your Mac using Apple's engines, ready to use right away.
- High: uses your cloud account for Speech recognition, Translation, and Summary, which is more accurate with regional accents, noisy rooms, and several speakers. If you don't have a connection yet, the app prompts you to Add account.
The Advanced section lets you choose each row separately: Speech recognition, Translation, Summary, Voice. For providers that support it, the Realtime streaming switch gives very low-latency transcription, but it's billed per minute of connection and needs a stable network.
iPhone and iPad
Go to Settings → Intelligence & engines, then tap Transcription, Translation, or Summaries. The picker screen has two blocks: ON-DEVICE (free, private, works offline) and CLOUD.
On iPhone and iPad, if you choose Cloud but the spoken language is supported by Apple at high quality, the app still uses Apple on-device for speed and no cost. The header of the recording screen always shows the engine actually running (On-device or Cloud).
Automatic engine switching when something fails
On Mac, if the engine you chose fails, the app switches to the default engine so the meeting isn't interrupted and shows a banner explaining what happened:
- Cloud transcription fails (3 times in a row, or the connection drops): switches to Apple on-device for the rest of the session.
- Translation fails: retries up to twice, then translates with Apple.
- Summary fails: tries the default engine, and finally creates a quick extract.
- Cloud voice fails: reads with the on-device voice.
On iPhone and iPad, if the Cloud connection fails when recording starts, the app switches to Apple on-device without telling you. If Cloud drops in the middle of a session, the audio is still recorded but new text may stop appearing. When Cloud translation fails, a banner appears and the app doesn't switch to Apple.