we’ve been exploring Rovo for some time, and we would like to know if it’s possible to customize or influence the UI in some way. Apologies if this has already been addressed elsewhere, but we haven’t found a clear answer yet.
When creating agents using the non-code Rovo solution (UI-based), or for example when using the Issue Organizer Agent, there is a functionality where the UI gives you a list of selectable options (such as checkboxes), and you can confirm your selections. We were wondering if there is a way to replicate or achieve similar functionality using Forge Rovo Agents (Actions).
And also from a development perspective, what would be a “strong argument” to choose the Forge-based solution over the non-code GUI solution? What are the key advantages of using Forge for building Rovo Agents? (externals APIs?, the potential of using complete Jira/Confluence APIs for actions? custom 3P systems? data preprocessing?)
I’m not sure I follow your ask about “selectable options”. Are you asking to expose Forge-based actions to user-defined agents?
My advice about building agents is to always start with the user-defined agents first. These are usually faster to build, and more importantly iterate. Because the “UI” of an agent is effectively the text it can read and write, the prompt engineering aspect of an agent is critical to get right. We’ve seen some sophisticated Forge agents crippled by a bad user experience, simply because the prompt was poorly built. Also, if you can get to “good enough” via a user-defined agent, then it’s all the better.
You’ve already hit on some of the “strong argument” items, so to confirm and expand, Forge is indicated for:
External APIs
Complete Atlassian APIs (ie you need to express things that aren’t in the smaller and simpler set of out-of-the-box actions)
Custom 3P systems, as implicit Knowledge Sources
Pushing strict logic into faster and reliable code implementations (for example, the strawberry problem is trivial to solve in code, as would be the class of math and logic problems)
Higher level of control & governance over change to an agent. Forge-based agents can be under version control, provide logging, and have explicit access controls, all governed by an admin. They can’t be “cloned” by end-users and modified.
The use of Forge storage to augment Knowledge Sources. I have yet to explore for myself, but I can see how some agents would need a permanent sense of “state” that isn’t dependent on chat history, current product context (like being on a page), or even product state (like existing page content or field values).
We’re still learning more indicators, so it feels like the distinction takes practice, even if I can list some “strong arguments”.
thank you so much for your thoughtful response! We’ve really enjoyed exploring Rovo so far and appreciate the insights you’ve provided.
To clarify my original question: I was specifically asking whether it’s possible to customize the UI within Forge-based agents to mimic some of the features found in user-defined agents.
We’re curious if it’s possible to replicate this kind of interaction in Forge-based agents, where users can select something via buttons or similar UI elements, rather than relying purely on text-based inputs.
Also we’ve noticed the provided built-in messages to inform users of their progress (e.g., “Analyzing: Reviewing your question” or “Invoking … "). Is there a similar way we can inform users at different stages of the workflow, or is this something we would need to handle by structuring the workflow ourselves?
Anyway, we really appreciate your breakdown of the key advantages of Forge. We’ll certainly keep these factors in mind as we continue to explore both approaches.
it’s possible to customize the UI within Forge-based agents to mimic some of the features found in user-defined agents.
Here’s a quick example of what I mean:
Your screenshot indicates the use of the Issue Organizer Agent which comes “out of the box”. If you view the details about this Agent within the Rovo UI (select Browse Agents, find the Agent, and select View Agent from the … menu) you should see there’s no specific UI customization. Therefore, you should be able to copy the prompt into a Forge Agent and it will act similarly, except mainly that the knowledge source will only be a single workspace where the forge app is installed (e.g. Jira). In other words, the UI in the screenshot can be almost considered as “Rovo magic”.
Also we’ve noticed the provided built-in messages to inform users of their progress (e.g., “Analyzing: Reviewing your question” or “Invoking … "). Is there a similar way we can inform users at different stages of the workflow, or is this something we would need to handle by structuring the workflow ourselves?
You could try asking Rovo to do this in your prompt. For example, you could define your prompt as a workflow with a series of steps and various steps along the way could be instructions for Rovo to send a message to the chat. Here’s a ficticious example:
Analyze the request and extract keywords.
Inform the user of the keywords identified.
Find articles containg the keywords.
Clear the output and present a list of the articles found.
I hope this helps, but as with much about Rovo Agent development, it will be a case of trial and error.
You are right - we noticed there’s no specific UI customization, and we’re wondering if the behavior we’re seeing might be tied to the “out of the box” Rovo actions. Using these actions in the GUI works fine (we can use the “select”), but with Forge (or in the GUI without these specific actions), we haven’t had success replicating this. With Forge it might be possibly due to the “Update” and “Create” defined actions not currently functioning as expected?
Also, we’re still struggling to provide real-time user updates specially during repetitive tasks (jobs, e.g., processing multiple issues in a loop) as feedback only appears after all issues are processed. It is no big deal though as we are still “playing” with it.
Thanks again - we’re enjoying experimenting with Rovo!
