How to communicate token usage?

[vemonet]

Hi,

Agent workflows require using LLMs, which translate in token usage, usually a dict containing prompt_tokens, completion_tokens, reasoning_tokens, total_tokens. Some systems might want to communicate token usage to the client (e.g. so they can monitor costs, or optimize tokens per request). I could not find any standard way to do so apart from using the CUSTOM event

Is there a recommended way to communicate token usage with AG-UI? Would it make sense to have one, or is it considered out of scope?