If you build an Agent without llm=, it does not call a model. It uses a keyless default reasoner that emits one RuntimeWarning and returns the input unprocessed. That is intentional for keyless local iteration, but if your agent does nothing, check whether you forgot to wire llm=. For a real model, pass a controller (or make_local_reasoner() with the providers extra). So the symptom of input-in, input-out with no model traffic is almost always a missing llm argument.

Context: My Julep Agent runs without errors but does nothing: it just returns the input unprocessed. No LLM call seems to happen. What is going on?