“On-device” and “cloud” describe where a particular answer path runs. They do not by themselves prove that a whole call works offline, that every device supports the same model or that one mode always answers faster. Check the control available on your exact app version and device.
Match the task to the path#
MagicInterview documents device-model answers on supported devices and a signed-out on-device demo; its release explanation describes the feature. Server-based coaching, spoken practice and research use Pro under the subscription terms. A project may still contain sources you have chosen for the task. Read the privacy policy and subprocessors for the processing details of the services you use.
| Question | What to check on your device |
|---|---|
| Availability | Does the current app offer the selected local or server answer control? |
| Input | Is your own microphone or the permitted call source actually received? |
| Context | Which project sources can the selected route use? |
| Connectivity | Does the task continue if the network changes? Test rather than assume. |
| Result | Does the answer stay within your facts and arrive in time for your task? |
Run the same harmless prompt through the available modes and inspect the result yourself. A local demo is an introduction, not a device-wide compatibility or latency benchmark. If your task includes a call, use the Windows or Mac audio checks and confirm permission under the legal-use guide. Open your workspace only after choosing the route your device actually offers.
Keep a text starting point available#
If your device does not offer the local model you need, use the story-map worksheet without model generation or try the limited cloud preparation preview. The worksheet is a browser form; the preview sends your supplied text for server-based generation. Neither establishes offline voice support on your device.