Escalations and Agent Memory

Hi
Is the Memory Agent already available and operational? I can see the option, but I am unable to check its functionality in any way. Where are the scenarios (outputs) after escalation saved for Memory Agent ?

hi, @Andrzej_Ekiert Yes, the UiPath Memory Agent is available and working. It allows agents to remember facts from escalations and observations across runs. The outputs or scenarios after escalation are stored as key-value pairs in agent memory, saved separately during design and runtime

Can I find somewhere those key-value pairs? I was trying to run it in debug mode and I was runing same case many times and I always have escalation. It looks like Agent Memory do not work at all. I was testing it from orchestrator and still nothing. Did you manage to truely test and run it? Are you able to provide some examples?

Enable Agent Memory in your escalation setup, set up the shared folder, and deploy your process from Orchestrator not debug. During runtime, the agent stores and reuses escalation data, which you can see in the Action Center.

Ex: Ask “What is my vacation balance?” during the first run..escalation happens.

Resolve escalation and save.

Next time you ask the same, the agent recalls the answer and doesn’t escalate

Hey,
Thank you very much for your input.
I’m wondering where can I find the escalation results which the agent uses? Can they be edited? What happens if a user responds incorrectly? The agent will not escalate each time, but will respond incorrectly. Is there any way to change these responses?

Escalation results are stored in Agent Memory and visible in Action Center but can’t be edited directly. If a user answers wrongly the agent might give wrong replies next time instead of escalating. You fix this by manually correcting or re-escalating tasks in Action Center. Improving prompts and using evaluation tools helps make responses better over time.

@Andrzej_Ekiert

Agent memory is not exposed to edit

it has a life time of 3months to start with…We can force escalation for those scenarios using prompt to reset

cheers