Long running Agent

[jamesespichanucsp]

Hello

I want to confirm my understanding of how AX behaves when deployed on Kubernetes.

Based on the GitHub documentation, my understanding is:

1. I deploy AX on Kubernetes.
2. I configure an agent in ax.yaml.
3. I forward the AX server port locally using:

kubectl port-forward -n ax rs/ax-server 8494:8494

4. I start a conversation using the AX CLI

My question is: if I close my local terminal or stop the kubectl port-forward, does the agent continue running inside Kubernetes?

And later, can I reconnect using kubectl port-forward again and run the AX CLI with --resume and --last-seq to review the outputs produced by the agent while I was disconnected?

For example

ax exec \
  --conversation d85a4b4e-c53b-4c84-b879-f10d905bce40 \
  --last-seq 2 \
  --resume

Is that the expected behavior?

[frosk1]

Yes, this is something that is very important and I think the docs and specs do not really define that. Can an agent here be seen as a long running service / process on substrate and thus a "task" in ax would mean monitoring the long running agent service process. Or is it more batch job like that each task has an input prompt with metadata on the agent env like resources, image etc. that is performed and when finished task is closed. Can it be both?

[frosk1]

Is AX more ment to be a system for one agent per user and not a multi user agent service model? Or is it just that the layer on top of ax and substrate is missing where each user requests would result in an ax task then