Usage Summary for realtime xai and Ultravox models

@livekitteams @LiveKit-Community
I’m trying to track real-time resource consumption in my LiveKit Agent app. When using Google or Azure, session.usage.model_usage provides a full breakdown of input/output tokens.

However, for xAI and Ultravox, the payload returns virtually empty metadata with no actual usage numbers

example:
Output: [LLMModelUsage(type=‘llm_usage’, provider=‘Ultravox’, model=‘fixie-ai/ultravox’)]

The specific metrics (like prompt_tokens or completion_tokens) are completely missing, unlike other models.
Is there a recommended workaround for this?

Those usage metrics are passed from the model providers. Unfortunately, Ultravox does not provide that information on their transport. So this is a question for their forum. If they add it, we would love to pass those value along to you: