Rovo Agents' Memory

[Sam Ben-David]

Does anyone have information on how the Rovo agents' memory function works.

For example - can we tell an agent to remember information based on feedback during a chat, and it's stored in memory? 

Or should this information be a knowledge source?

[Rebekka Heilmann _viadee_ · Accepted solution]

@Sam Ben-David 

Currently Rovo is not memorizing anything beyond the specific chat history. If you prompt feedback, the specific Agent chat will take that into consideration. That feedback is not retained for future interactions though.

The feedback function sends feedback directly to the Atlassian product team. It is not (yet) used to improve agents.

From what I understood at Team Europe though, that is something that will come in the next few quarters - most likely along features like versioning of Agents

[Dr Valeri Colon _Connect Centric_]

@Sam Ben-David Rovo agents don’t store long-term memory from chats. Anything you tell the agent during a conversation is temporary and won’t persist. If you need the agent to “remember” rules, facts, or preferences, add them to the scenario or a knowledge source. That’s the supported way to make the behavior consistent across sessions.

[Nikola Perisic]

Hello @Sam Ben-David 

Generally, Rovo can store a chat history like ChatGPT is currently doing. Even on this page: https://www.atlassian.com/blog/announcements/team25-europe-rovo-everywhere, you will find a quote saying: "Rovo never forgets'.

If you are using Rovo Dev CLI, you can check this out: https://www.youtube.com/watch?v=542NAf-KwjM