Currently, we only support shorthand (e.g. "openai::got-4o-mini") specification of models for functions passed to the agent.

vLLM models require a model config that specifies theapi_base, at least.