While working with an agent, especially if it includes tools, it is really useful to access details and metadata of execution steps. The bundle already tries to improve that situation with traceable decorators and an UI.

But I was wondering if we should introduce some kind of trace/protocol of the chain execution (incl. raw messages and potential tool details) out of the box.

MLFlow solves that quite impressive:
https://mlflow.org/docs/latest/tracing/