Oracle Fusion AI Agent Studio: My First Hands-On RAG Agent

Oracle Fusion brings a lot of excitement around embedded AI, but what does it take to actually build something functional? In this post, I take a practical look at Oracle AI Agent Studio by building my first hands-on Retrieval-Augmented Generation (RAG) agent from scratch.

What is AI Agent Studio?

Oracle AI Agent Studio is a new capability within Oracle Fusion Applications that extends HCM, ERP, and SCM with built-in AI agents. While it was first introduced in the 25C release, it has evolved rapidly with significant enhancements in subsequent updates.

Because it is native to Fusion, the platform can seamlessly access enterprise data while respecting existing security models and role-based access controls. Its low-code approach makes it accessible to a broad range of users—whether you’re configuring agents from pre-built templates or building them from scratch using available tools.

Oracle describes this approach as “built-in, not bolted on,” which is fitting. Since the capability is embedded directly within Fusion, it enables AI agents to operate on enterprise data in a secure, governed, and context-aware manner—without the need for external integrations.

Use Case:

In this exercise, I explore Oracle AI Agent Studio by building a simple RAG-based document agent. I will use the pre-built HCM template, Benefits Policy Advisor, as the starting point.

Configuration:

Access to AI Agent Studio and runtime agents can be controlled through custom roles at three levels: User, Developer, and Admin. For more details on setup and security considerations, refer to this great blog from Oracle AI COE: Securing Oracle Fusion AI Agent Studio: Setup & Access Guide | fusioncoe

For this exercise, I will use the Developer role. After logging in to Oracle Fusion, I will navigate to AI Agent Studio from the hamburger menu by selecting Tools>AI Agent Studio.

The landing page displays all published templates, including both pre-built and custom options, under the AI Agent Studio tab. In release 26B, the page has a familiar GUI layout with additional options at the bottom. For more information, refer to this blog: Learning Path for Fusion AI Agent Studio | fusioncoe

For this RAG agent use case, the setup follows this hierarchy:

Agent Team -> Agent -> Tools -> Document

I will create an agent team and use the existing worker agent, Benefits Policy Advisor.

1. In the Agent Teams area, click the + symbol to create a new agent team. Enter the name and code as shown below, and choose the supervisor type.

For this exercise, I will review three other tabs:

• LLM: I will go with the default model, GPT-4o mini.

• Security: If we want to expose this to specific roles, we can add those roles here. I will choose “For all employees.”

• Questions: I will enter three starter questions related to benefits policy.

2. After saving it, from the left panel, choose “Benefits Policy Advisor” from Available Agents. Since we are using pre-built agents, we will look under Available Agents.

   

   
   

3. Save and Publish the agent team.

Next, upload the policy document to the Document tool. To show the impact clearly, I will test the agent both before and after uploading the document.

There are two ways to run an agent. You can run agent teams from Explorer or by adding them to a Redwood Guided Journey. For this exercise, we will use the Explorer view using this path.

Home > Me > Show More > AI Agent Studio >AI Agents. 

I look for my agent team and click it to Run. It opens up with the three starter questions I provided when defining my agent team.

I clicked the first question, and the answer was very generic one from LLM training, rather than a specific document.

To further verify, we can test it directly from Studio. Testing in Studio provides detailed insights into what’s happening behind the scenes.

In the Agent Team Edit area, click the Play button in the top left corner.

Then select the first question to view the response below.

The document tool lookup_benefits_policies is triggered as part of the Agent’s workflow. Click the <> button to view the calls that were made. As shown, no documents were returned.

Now comes the important step of enabling it as a RAG agent. We attach the pdf policy document to the document tool and publish it.

Upload the Document :

I looked up the tool from the definition: lookup_benefits_policies.

I go to the Tools area and open this tool. I upload my test policy PDF document, which starts in draft status.

I change the status to “Ready to Publish” and save.

4. To make this document available to the agent, next step is to run Fusion ESS job “Process Agent Documents”. You must manually run this ESS job and on successful completion of that job, the status of the Document in AI Agent Studio changes to "Published".

Last step is to republish the agent team. So open the Agent team and click Publish

5. Now I will test it again from the Explorer view by running the agent after adding a document to the Document tool. You can clearly see that the responses are grounded in the uploaded document, with source references from the document, confirming that it worked.

Lets test it again from the Studio to see details. we start by going to Studio, lookup our agent team and click the Play button in the top left corner. Select the first question to see what happened.

Clicking the <> button next to the lookup_benefits_policies document tool confirms that the agent can now locate the policy document and use it to generate the response.

Query Scenario	Source of Truth	Answer Quality
Before Document Upload	General LLM Training Data	Generic, high-level, and not aligned with your organization’s actual guidelines
After	Uploaded Policy Document (RAG Source)	Response grounded in the policy document, along with exact source references.

Final Thoughts: This exercise helped me build and run my first AI agent—a simple RAG-based agent—in a short amount of time. The process was straightforward and intuitive, which makes me more confident about expanding it further with additional tools. As Oracle continues to enhance this functionality, AI Agent Studio is becoming a strong starting point for anyone looking to explore AI agents in Fusion. I highly recommend giving it a try.