In short
A configured offline model can recognize speech on your PC when you have no connection or the recognition service is unreachable. If cloud credits run out, you can explicitly choose local dictation. While online with credits, Zack Typing continues to use the cloud.
Prepare before you travel
Sign in while connected. Under Advanced → Offline dictation, enable offline dictation, download a model and select Use. Test a recording without a connection before relying on it.
Small is included with a free account and supports English, German, Spanish and French. Medium and Large are available from Basic and cover more languages. Larger models need more storage and processing power.
Check the language
Small does not detect the language automatically. It uses the first supported language in your selection, or German without a suitable preference. The app shows the active language beside the model. Medium and Large offer automatic detection; see multilingual dictation.
What works locally
Offline speech recognition uses no credits. Its results are marked as transcribed on this device. Local history also remains accessible.
Chips and automatic dictionary learning are separate model tasks. They do not run in the built-in offline speech model and need a connection in the standard Zack configuration.
Downloading a model does not force local processing
The setting prepares local recognition for the cases above. It does not prevent cloud requests while you are online with credits. Choosing Transcribe in the cloud later also sends a previously local recording for cloud recognition.
Related questions
What if no offline model is available?
A retained recording can be transcribed later when the connection works again. Configure the model and language beforehand if you need text immediately while away.