Choose models for your service

Before changing models, identify the organization and the assistant you are configuring. Open Models in the application. Organization defaults apply unless the assistant has a selective override in Settings → Model Overrides.

Decide who operates the models

With BYOK, you supply the selected providers' credentials. Configure listening, reasoning and speaking separately, and add an embedding model when using document retrieval. Check each provider connection before running a conversation.

With IntraCord managed models, use an IntraCord Service Key and the available voice, speed and language settings. Service Keys authorize model service usage; application API keys authorize access to the integration API. Keep these purposes distinct when choosing a credential.

With Speech to Speech, a realtime model handles spoken input and output. Configure the supporting LLM for extraction and quality analysis as well; document retrieval still needs an embedding model.

Test the language that customers actually use

Use names, street addresses, numbers and interruptions from your own service. Check recognition accuracy, pronunciation and response delay together. An assistant-level override lets you change one service while retaining the organization's other settings.

Check where information goes

Your selected model endpoints determine where audio, prompts, conversation history and tool results are processed. Running the application on your own server does not keep data local when you use external model endpoints. Review data destinations and the local Hungarian voice setup when planning an on-prem installation.