In short
With your own API key, Zack Typing sends recordings directly to your provider. Choose a text model as well for chips, titles and other text tasks. The two tasks can use different providers, without consuming Zack credits.
Connect a provider
- Under Advanced → AI provider, choose Own API key.
- Add your provider in Models and enter its key.
- Choose the provider and model for Speech recognition, and for Text editing if needed.
Two model slots with different jobs
Speech recognition handles dictations and imported audio. Text editing handles chips and other text tasks. Without a text model, dictation still works, but those text actions are unavailable.
A chat model is not automatically a speech model. Use a provider route offered for the task. Speech-recognition presets include OpenAI, OpenRouter, Groq and Mistral.
Check capabilities with a short recording
Try a sentence containing one of your dictionary terms. Some providers and models have limited support for custom instructions or preferred spellings. Zack Typing reports when those hints could not be applied.
The app accepts imported files up to 40 minutes and 64 MB. Your provider may have a smaller limit for the transmitted request body. Prices, quotas and billing for failed requests also follow its terms.
Local servers and switching back
Configure a compatible local server through its base URL. Ollama and LM Studio are text-model presets; that does not establish speech-recognition support. Speech needs a suitable endpoint or the built-in offline models.
Switch AI provider back to Zack Software whenever needed. Your own providers remain saved for later use.
Related questions
What should I check if the key is rejected?
Check that the key and base URL belong to the same provider and that the key is valid. Enter it again and use Check. Change one setting at a time and test briefly so you can identify what helped.