Answers and transcription are separate choices
Open settings and find AI Providers. The connection that generates answers is separate from the one transcribing audio. Set the provider, endpoint where applicable, key, and model for each.
The backend includes adapters for OpenAI and other services, as well as compatible endpoints. A response provider does not necessarily support transcription or live audio.
Verify the connection
Use the testing and catalog controls available in the app. The model list depends on your key's permissions and the provider response; some services do not expose a compatible catalog.
Check the capabilities advertised by your environment before starting a session. In cloud deployments, the operator may restrict destinations and providers.
Live audio
Live audio has its own connection settings. Gateway adapters include OpenAI and Deepgram, depending on backend-enabled capabilities.
Select microphone and system audio sources. Permissions and capture availability differ between the browser and desktop.
Costs and credentials
APIs may charge by tokens, audio minutes, or other usage. Local installation does not remove those charges. Check your provider dashboard and set spending limits.
Keep keys in the intended configuration flow. Do not put them in URLs, public code, demos, or shared screenshots.