The Centrl Agentic Platform, CentrlX, helps your team move beyond basic chat to automate complex workflows using governed agents. From the Agents page, you can now create new agents, configure their instructions, and connect agents to different data sources. This article explains how to create an agent, update the configuration, manage versions, and reuse or share agents to consistently generate rich, review-ready deliverables like slide decks, spreadsheets, comparison tables, and reports.
This article will cover the following topics:
If you need assistance with creating and managing AI agents, you can find more details in the article below:
If you are looking for information regarding the Research Assistant chat feature, please see the dedicated article below:
You can start an agent from a new chat in the Research Assistant.
-
Navigate to the Research Assistant module and select the Agent tab
-
Select Start Flow or click Agents in the chat box.
-
Search for the agent you want to use to begin.
-
Depending on the agent configuration, you may be prompted to include files or use other connectors.
Once you select an agent, it starts a dedicated conversation thread so you can complete the workflow in one place.
Additional Notes:
-
The Agents available to you depend on the options configured by your organization, and access and permissions for your user profile.
-
Agents execute fully asynchronous workflows that typically run for a few minutes depending on complexity. Agents focus strictly on document and deliverable creation.
After you start an agent, it may ask clarifying questions before it begins working. This helps the agent better understand your goal and produce a more useful result. Agents may ask for ad hoc files or user credentials to user-specific connectors like Google Drive, depending on what connectors are enabled for the agent.
Depending on the agent, you may also be prompted to provide files. These files can come from:
-
Direct uploads
-
Connected file sources
If the agent supports file inputs, add the requested documents and continue through the workflow.
Once you have selected your files, you will click Submit. If there are no files to provide, you can click Skip.
Note: File prompts only appear for agents that are configured to use them.
Agents can use connected systems and data sources to gather information while completing a task. Depending on how an agent is configured, it may use:
-
Platform data: Accesses targeted data from Partners, partner products, attributes (such as AUM, cyber security ratings, and last audited dates), assessments, and related files.
-
Web content: Utilizes a modernized agentic web search engine. This process takes a few minutes but provides significantly more analysis depth than standard chat web lookups.
-
Files from connected repositories.
-
Email content and attachments: The Outlook integration can read emails and automatically draft email replies in your inbox based on agent output. It cannot directly send emails; users must review and manually click send from Outlook.
If a connector requires your authorization, you may be prompted to authenticate before the agent can use that source. Once access is granted, the agent can continue using the approved connection as part of the workflow.
For example, if the Agent is configured to use the Outlook connector, you would be prompted to connect to your Outlook account:
If you are connecting to the integration, proceed through the process of connecting and click Finish. If you do not need to connect, you can click Cancel to proceed.
Note: A Reset Credentials utility is available to remove old connection credentials and prompt re-authentication if an integration runs into authentication errors.
Note: Available connectors depend on your organization’s setup and the design of the selected agent.
You can find more information regarding the integrated AI agent connectors in the article below:
Some agents provide an Execution Plan after the conversation begins. This plan outlines how the agent intends to complete your request and helps set expectations before the final output is generated.
The plan may help you understand:
-
The major steps the agent will take
-
How the task is being organized
-
What the agent is focusing on during the run
As the conversation becomes more specific, the thread name may also update to better reflect the task you are working on.
Note: The amount of detail shown in the plan can vary by agent.
When an agent finishes, it can return results in different formats depending on the workflow.
Common output types include:
-
A formatted response in the conversation.
-
A rich text result for easier review.
-
Downloadable files or generated documents (fully supporting Microsoft Word documents, PDFs, PowerPoint presentations, or Rich Text formats).
For rich text results, you may see options to preview or download the content. If the agent creates files, they are typically displayed directly in the conversation so you can review or download them as needed.
This makes it easier to move from research to a usable deliverable without leaving Research Assistant.
After an agent returns a result, you can continue working in the same thread by asking follow-up questions.
Follow-up prompts can help you:
-
Refine the original output.
-
Ask for clarification.
-
Request another version of the output.
-
Continue working with the same context.
This allows you to build on the completed work instead of starting over in a new conversation.
Note: Follow-up questions are most effective when they stay focused on the same task for output.