[Feature request] Session context for holmesGPT

[pavneeta]

Today, holmesGPT CLI treats each CLI command as standalone and does not retain context such as which cluster, which cloud provider, and sub - resource context ex: subscription ID , resource-group etc. ; this means that while it is doing agentic iterations and loop; it is not able to have a back and forth troubleshooting engagement with the user

Example:

Feature ask:

1. Retain the environment context in memory for the duration of a session (time based, or terminal session based) where it stores and can recall whcih resources are being talked about
2. It retains the previous outcomes/responses/checks done of its runs and user prompts so that it can reduce user input needed in prompts (in trying to rehash the same context), and can diagnose more complex issues which require a few iterations between user and the agent

[arikalon1]

Hey @pavneeta

@lightme16 created #395 that allows running an interactive session, with historical messages saved during the session. Does this solves this issue?

[pavneeta]

https://github.com/robusta-dev/holmesgpt/pull/395 is a step in the right direction, it adds context of the resource and creates a one-step clarification session. I envision a session ttl == till the customer query/issue is resolved manually (ack! ) or the terminal session itself is terminated, where the session has a memory so that you can go back and forth, where the agent remembers what it has done so far , for example, below : iI want to be able to go do 1, 2, or/and 3

[lightme16]

@pavneeta hello, did you try --interactive command? eg holmes ask "what pods are unhealthy and why?" --interactive It would run interactive loop(code) and you can chat with holmes after your session?

[aantn]

@pavneeta closing, as I think --interactive covers this. If not, let me know and I'll reopen.