Knowledge sources for Rovo agent scenarios
Knowledge, in the context of agents, is effectively all the source information the agent can use to answer user prompts. Knowledge can include links to Confluence and Jira spaces, individual pages, Google Drive workspaces, and other relevant resources.
Every agent has access to all organizational knowledge unless you specify otherwise. This is effectively the organization-wide resources that your org administrators have connected via Teamwork Graph and Rovo.
If you configure an agent with custom knowledge, you narrow the scope of knowledge by specifying which resources it should look at. Although this may sound limiting, it can make responses more accurate and helpful by focusing the agent on information most relevant to skills it was built to perform.
No matter the knowledge source, agent’s won’t return answers with information from sources the user doesn’t have clearance to view. It respects the permissions of the user who prompted it because it is acting on their behalf.
Deep research mode
If you choose to customize the type of knowledge your agent has, you can turn on Deep research mode to enhance how your agent “thinks” about that information. This is similar to selecting Deep research mode in Rovo Chat, effectively turning your agent into a research assistant. It will consider user prompts more broadly and deeply. And it will respond with comprehensive, reports that follow the formatting, tone, and structure you outline in the agent instructions. This mode is most useful in scenarios where your agent needs to conduct heavyweight evidence-based research.
Considerations for Deep research
If you customize the knowledge of a specific scenario with Deep research, the agent will respond in Deep research mode any time a user’s prompt triggers that scenario.
By tying an agent’s research capability to specific custom scenarios, you control which questions trigger this mode. This differs from Rovo Chat, where users select the mode as part of their prompt.
You can’t add Deep research to the Default scenario because research mode is an exception to be triggered when deeper thinking is warranted.
Agents with deep research–enabled scenarios can also be triggered in automations, allowing agents to generate in-depth research reports as part of automated workflows or scheduled processes.
However, before you add Deep research, consider this:
Deep research mode takes longer to respond — up to 15 minutes — while it analyzes and compiles an in-depth report
Automation currently times out after 15-minutes, so if deep research runs for longer the automation will fail
Each user or automation is limited to 30 deep research requests a day
Add knowledge to an agent’s scenario
An agent’s scenario contains the trigger, instructions, knowledge, and skills that guide it in specific situations. You can add more than one scenario.
When it comes to configuring knowledge specifically, open a new or existing scenario and scroll down to the Knowledge block.
To configure the knowledge scope and thinking mode:
Select one the following:
All organizational knowledge (everything the organization allows)
Custom knowledge (specific resources you choose)
No organizational knowledge
To add custom knowledge sources to an agent’s scenario:
Select Custom knowledge
Select the Add knowledge button
Select the checkboxes of sources you want your agent to access
(optional) Toggle on Deep research mode
If you select Confluence, Jira, or Jira Service Management, you can focus the scope even further to specific spaces — or to a specific branch in a Confluence content tree. Selecting Content under will include the parent content with all child content items.
Skills and Knowledge
Agents use skills to retrieve and process information. There are some system skills that are tied to knowledge because, when you provide custom knowledge to your agent, we’ll customize the system skills your agent uses to fit. When you create an agent, you’re unable to configure the skills your agent can use.
Skills allow agents to access data from your other Atlassian apps and connected third-party apps in order to provide their full functionality. For example, while you may use the Work Item Organizer in Jira, it will be able to leverage information based on user permissions from Confluence pages to create a new epic in Jira.
These are the system skills your agent will have depending on the knowledge sources you provide it:
Name | Relevant knowledge | Description |
Content read | All sources | Retrieve the contents from certain URLs including Confluence, Jira, Google, SharePoint, Figma, GitHub, Loom, Microsoft, Confluence Whiteboards, Slack, and others. This also includes the current browser URL. |
People | All sources | Fetches the account ID, email, name, location, job title, department/organization, recent activities, and profile picture based on a user's name. Do not call this function with a job title or team. |
Search | All sources | Search for information from all organization knowledge sources or custom knowledge sources including Confluence, Jira, and third parties. It returns relevant content and sources that were used to find the question. |
Page search | Confluence sources | Finds a given Confluence page or blog post based on certain search filters. |
Jira field search | Jira sources | Search Jira for fields, their values, and functions in JQL expressions. |
Jira work item search | Jira sources | Search Jira work items using a JQL expression. |
Jira space search | Jira sources | Search Jira spaces based on user query. This tool can get you space details like space key, ID, title, and type. |
Jira recent space | Jira sources | Get the user's most recently used Jira space. This tool can get you space details like space key, ID, title, and type. |
During the creation process if you select certain skills or knowledge sources for your agent, we’ll customize the agent’s skills to suit them. For example, if you create an agent that can create Jira work items, or point it to a Jira space as a knowledge source, we'll enable the Jira field search skill for that agent, so it can retrieve information from Jira efficiently. We do this to improve the performance of your agent. If the agent is attempting to use too many skills in response to a prompt, it can slow down performance.
Was this helpful?