What are the memory limitation of Rovo's agent during a long chat?

[Seb Blandin]

I'm currently building a Rovo agent for knowledge transfer. We have 44 questions to be asked by the agent to the user before publishing a page summarizing all his answers, however when the user answer to a bunch of questions, the first answers are not retrieved by the agent when it's time to publish the page. As a workaround to this short term memory, I asked the agent to publish a summary of the answers before asking the next question. A colleague tried to answer all the questions, but on ten sections only the last seven could be published as the Agent would display for the first three : summary as above. So when he wanted to publish the page, we would have summary as above as answers to the questions for the first three sections. Any idea on how to prevent this behaviour?

[Rebekka Heilmann (viadee) · Accepted solution]

Feels like it's not a great use case for an Agent. Sounds like you need the Agent to 

A) always ask the same question in the same way

B) in the correct order

C) AND remember all results correctly

NOT a great use of GenAI.

How about:

1. Build a form with the 44 questions. -> This makes sure it's always the correct questions in the correct order and answers are obtained.

2. Give that form to a Rovo agent

3. let the Rovo agent summarize the answers and publish to a Confluence page

Step 2 might be a bit tricky but is achievable depending on how you implement your form.

[Seb Blandin]

Thanks a lot Rebecca for your feedback. That's what I thought, that we shouldn't use agent for this kind of use case. Just as matter of curiosity, What would you use to build the form?

[Rebekka Heilmann (viadee)]

We got Microsoft Forms for that kind of stuff but you could use anything that gives the results as csv. Pretty sure Rovo can understand that if attached to the Chat.

Worst Case you can create a Confluence table from the csv.

Rovo can then go through each row and create Confluence pages from it - not sure if that's one step or it needs to be iterated. That could be done through an Automation though :)