Answer engines¶
The engine is the language model that writes the answer. Pick it with the Engine control: in the sidebar on iPad and Mac, and in Settings on iPhone and iPad. You can change it between questions. The chat remembers which engine wrote each answer.
| Engine | Where it runs | Needs |
|---|---|---|
| On-device | This device | Apple Intelligence on |
| Private Cloud | Apple's Private Cloud Compute | Network; iOS, iPadOS or macOS 27 |
| Worker | A server you run | A worker address |
| Passages only | Nowhere: no answer is written | Nothing |
On-device¶
The default. Apple's built-in model runs on the device, so nothing is sent anywhere and it works in airplane mode.
Its limit is size. It reads about 4,000 tokens at once, which is a handful of passages, so its answers are shorter and plainer than the other engines'. Narrowing the scope helps it most.
If Apple Intelligence is off, or the model is not ready, the chat says so in the line under the title, and each answer area gives the reason. The search and the passages still work.
Private Cloud¶
Apple's larger model, on Apple's Private Cloud Compute servers. It reads about 32,000 tokens, eight times as much, so it gets about twelve passages and writes better prose. Your question and the chosen passages go to Apple under Apple's privacy guarantees. They do not go to us.
Two costs:
- It needs a network connection.
- It draws on your iCloud allowance. When you near the limit, the engine settings show a usage line and a Show usage options button that opens Apple's own sheet.
Apple's content filter can stop a Private Cloud answer partway, and no setting turns it off. See Troubleshooting.
Worker¶
A server you run yourself, such as a Mac with a large model on it. Set its address under Worker in Settings. Then pick the provider (oMLX, Ollama or OpenAI) and a model. The app has no default worker and contacts none until you enter one. It is also the only engine that can draw illustrations.
Passages only¶
No answer is written. You get the search results to read yourself, which is faster and has no model's interpretation in it. It suits a question where you want the text, not a summary of it.
In-app help¶
Settings, then Help, explains these engines and every setting in the app, in the app. On a Mac, the Help menu has The Knowledge Press Help. The question-mark button beside the Search, Answers and Synthesis headings in Settings opens the matching page.