Skip to main content
When the AI processing step (the language model that formats your text) fails on macOS, Superwhisper pastes your raw transcription instead. You see unformatted text where you expected the mode’s output. Your words aren’t lost: the dictation is saved in History, and you can reprocess it once the cause is fixed. On Windows, you may see an error instead. The messages are Error during local LLM inference or Error while making chat query. The fixes below apply to both.

​
Fix the error

1

Check your connection

Most language models run in the cloud, even in modes with a local voice model. If you’re offline, switch the mode’s Language Model to a local one like S1-mini or reconnect.
2

Retry from History

Open the dictation in History and reprocess it. Transient provider errors clear on retry.
3

Switch the language model

Edit the mode and pick a different Language Model. If the error follows one specific model, that provider is the problem.
4

Update Superwhisper

Use Check for Updates in Settings → Configuration. Updates can include fixes for provider changes.

​
Errors with your own API keys

If the failing model is one you added under Settings → Models library → Bring your own keys:
  • Re-check the key: paste it fresh from the provider’s dashboard. Expired or restricted keys can fail.
  • Google Gemini keys: if a Gemini key keeps failing, create a new key in Google AI Studio and paste it fresh.
  • Check provider status: an outage at the provider can make the AI step fail.
Built-in cloud models need none of this. If you don’t need your own keys, switch the mode back to a built-in model.

​
Report it

Turn on Error logging in Settings → Configuration, reproduce the error once, then report the recording from History. Name the mode and the model it uses in your report.