Have you considered adding a run_id to all messages and not just the RUN_.* messages?

[marcushill]

I'm finding my way around the open AI ecosystem and I stumbled on this protocol. It seems like having a run_id to all events would allow an arbitrary client/server to multiplex several different agent runs over a single connection without having to negotiate a different method of separating streams at the protocol layer.

[mme]

@marcushill

That's interesting and I did consider it. However, I didn't want to complicate the protocol before building an actual prototype of the multiplexing you describe.

Is this something you are interested doing?

We can get a branch going with run_id per event.