Speaker A
Aditya, and today we will be taking a look at how to use Oracle integration to power Fusion AI agents.
Learn how to use Oracle Integration with Fusion AI Agent Studio to create intelligent agents that access business data and external systems securely.
Key Takeaways
- Fusion AI Agent Studio integrates natively with Fusion Cloud for secure, seamless AI agent creation.
- MCP tool simplifies connecting AI agents to external systems without extra development overhead.
- Oracle Integration Cloud enables powerful data lookups and interactions with on-prem and cloud databases.
- Agents can be customized with specific prompts and tools to handle business-specific queries effectively.
- Pre-built adapters and flexible integration options unlock extensive connectivity possibilities.
Summary
- Fusion AI Agent Studio leverages large language models to create agents that reason, plan actions, and interact with users.
- Agents can call various tools including business object, document, user profile, calculator, email, deep link, and MCP tools.
- The MCP tool enables secure connection to external MCP servers without building additional REST wrappers.
- Demo shows integration with an on-premises claims database accessed via Oracle Integration Cloud (OIC).
- Integrations use input parameters like EMPID to query databases and return JSON payloads with relevant data.
- Agent Studio is accessible within Fusion Cloud applications for users with appropriate roles.
- Users can add active integrations as tools in Agent Studio and configure MCP server connections with client credentials.
- Agents are created and managed within agent teams, with prompts defining their responsibilities and tool usage.
- The agent can fetch user session data to personalize responses based on the logged-in user.
- Oracle Integration offers 100+ pre-packaged adapters and technology adapters (REST, SOAP, FTP) for flexible system connectivity.
Full Transcript — Download SRT & Markdown
Speaker A
A quick intro on what Fusion AI Agent Studio is.
Speaker A
What the agents do is they leverage a large language model to reason, make action plans, and interact with human employees to gather information and take direction and basically make a tool call.
Speaker A
The Agent Studio provides a design-time environment that provides a set of tools to create, customize, validate, and deploy all of these agents.
Speaker A
This is built natively into Fusion Cloud applications, so when you go into your SAS environments, if you have the correct roles, you will see under the tools that you have the Agent Studio option there.
Speaker A
And then also, because this is natively built into Fusion Cloud, this provides secure access to the Fusion knowledge stores and the APIs internally.
Speaker A
So when you create an agent, it leverages a large language model.
Speaker A
So an LLM to talk to whoever is interacting with the agent to understand the question.
Speaker A
And then take direction on which tool to call based on the question asked.
Speaker A
And that's why there are a set of tools that are provided when you create an agent.
Speaker A
And some of these tools are the business object tool, which provides access to the Fusion knowledge stores.
Speaker A
All the underlying business objects within the Fusion applications, you can access those.
Speaker A
And again, since this is native to the Fusion Cloud, you're not making a call outside.
Speaker A
There's a document tool, in this case, in this example, we'll take a look at that as well.
Speaker A
But you can upload a document and then the agent can then scan through the document.
Speaker A
Get you an answer based on the question asked.
Speaker A
Some other tools are user profile, calculator, email, deep link tools.
Speaker A
And the one we're interested in here for this demo is the MCP tool, with this tool you can securely connect to external MCP servers.
Speaker A
Use their capabilities inside of the agents without needing to build additional rest wrappers.
Speaker A
So now moving on to the demo portion for this video.
Speaker A
There, let's just assuming that there's a database that's on-prem or private in another cloud.
Speaker A
And as a part of the interactions with the agent, it requires to look up some details in the database.
Speaker A
In this case, particular example, the agent is a benefits agent that has access to things like the documents.
Speaker A
The enrollments business object.
Speaker A
And then also, when in cases where you're enrolled with some provider.
Speaker A
And you make claims as an employee.
Speaker A
Their claims are stored inside their systems and their databases.
Speaker A
And a user or an employee here inside of Fusion is trying to ask for the list of claims that were made in this year.
Speaker A
And when and when the agent gets that question.
Speaker A
It's going to invoke an integration call.
Speaker A
And the integrations is then going to go and look up the claims database and return a list of claims.
Speaker A
That were made for this year for that user.
Speaker A
With integrations, you can unlock endless possibilities in the sense that there are 100 plus pre-packaged adapters.
Speaker A
To create integrations to these systems.
Speaker A
And if there isn't an adapter for that particular system that you're trying to access.
Speaker A
You can always use the technology adapters, you know, rest, soap, file, FTP.
Speaker A
To talk to these systems, either post or get information.
Speaker A
Okay, so let's take a look at the integration that I am going to call from my agent.
Speaker A
This integration is accessing an on-prem database and getting a list of claims back.
Speaker A
The input parameter to this is the MPID that is available within the Fusion environment.
Speaker A
Which will be passed in, and based on that EMPID, does a lookup and returns a list of claims in a JSON payload.
Speaker A
This is the endpoint URL.
Speaker A
And this is the response sample.
Speaker A
Okay, next, let's go ahead and add this active integration as a tool.
Speaker A
So in the menu inside of the projects, you'll see now there's a new icon for the agents.
Speaker A
And in here, there's a section for tools, and I will click this plus here.
Speaker A
And it will bring me all the list of the integrations that are active.
Speaker A
And I wanted to select this one.
Speaker A
Also pulls in the description that I entered when I created the integration.
Speaker A
And then I can enter in optional guidelines to give it guardrails, description.
Speaker A
That will help the LLM understand what this field is.
Speaker A
But in this case, it's all self-explanatory, so I'm not doing that.
Speaker A
I'll click on save.
Speaker A
And the tool is created successfully.
Speaker A
Now, I have an MCP server already enabled.
Speaker A
But in case you haven't, you click that pencil.
Speaker A
The menu opens up, you click this to enable the MCP server.
Speaker A
And you get the URL from here.
Speaker A
Next, let's take a look at the Fusion AI Agent Studio.
Speaker A
I'm in my Fusion environment, under my tools.
Speaker A
I can see the AI Agent Studio option because I have the correct roles.
Speaker A
And when I click into it, it'll load.
Speaker A
The AI Agent Studio homepage.
Speaker A
At the bottom, you'll see all the options or components of the Agent Studio.
Speaker A
The agent teams, agents and tools.
Speaker A
And then, you know, some more options to configure the agents.
Speaker A
So what we're interested in during this demo is the tools section.
Speaker A
Okay, so let's add a tool.
Speaker A
I'll select MCP.
Speaker A
Call this OIC MCP benefits.
Speaker A
I'll select the family as common.
Speaker A
Others.
Speaker A
And in description, OIC MCP tool.
Speaker A
And for now, I won't select this, but you can add that.
Speaker A
And in here, we need the instance URL.
Speaker A
So we'll go back to the OIC page, copy the MCP URL.
Speaker A
Come back here.
Speaker A
Paste that.
Speaker A
The transport type is streamable HTTP.
Speaker A
And the credential type is going to be client credentials.
Speaker A
Now, how to get the token URL and client credentials, I'm sure there are multiple videos out there explaining how to create a confidential app.
Speaker A
In OCI IAM domains to get the token URL and the values for the client ID, secret and all those things.
Speaker A
So, let me just copy over the format.
Speaker A
And basically, you need to give a JSON payload here with four keys.
Speaker A
Grant type, client credentials, scope, client secret, and client ID.
Speaker A
And all of these values, of course, you can get from that confidential app that you create in the OCI domain.
Speaker A
Except for this, it's hardcoded client credentials.
Speaker A
Once you enter these details, you click on update.
Speaker A
And what this will do is this will go and get the list of available tools on this MCP endpoint.
Speaker A
And here, in itself, they have a playground, so I can test things out if I wanted to.
Speaker A
So let's go ahead and do that.
Speaker A
I'll just type in EMPID.
Speaker A
I know that's the parameter they showed the detail here up as well.
Speaker A
And I can click on preview.
Speaker A
And this will go and fetch the details.
Speaker A
Meaning everything's working good.
Speaker A
So I'll click on cancel.
Speaker A
I'll select this one, this one I don't really want for this use case.
Speaker A
So you can just ignore that.
Speaker A
And then you can click create and it'll create the MCP tool for us.
Speaker A
Next, let's go to my agent teams.
Speaker A
And I already have an agent team for the benefits advisor.
Speaker A
And I'll open this, the one that I've created.
Speaker A
And you'll see here, in the agent teams, you can do it two ways.
Speaker A
You can create an agent, add a tool to the agent.
Speaker A
And then in the agent team, come and add the agent.
Speaker A
But here, I've just directly went into agent teams.
Speaker A
And created the agent here.
Speaker A
Right, so if you take a look at this agent that I have.
Speaker A
It is a worker agent that I've added.
Speaker A
I've given it a name.
Speaker A
And then the prompt here, right.
Speaker A
This is what it actually takes in the AI agent to understand what it needs to do.
Speaker A
What its responsibilities are.
Speaker A
How to identify what tool to call based on what information.
Speaker A
And all those things.
Speaker A
Right, so here I have for claims, any sort of claim question.
Speaker A
You know, call this tool to receive claim information.
Speaker A
And then at the start, I have given it a responsibility as well.
Speaker A
Then the tools for my agent here.
Speaker A
Are, you know, a lot of different tools.
Speaker A
The main ones being the get user session, so whoever is talking or interacting with this agent.
Speaker A
It can get the user session.
Speaker A
And this basically, if you look at the details, it just fetches the person number of the logged in user.
Speaker A
This comes in handy when you need to do more lookups.
Speaker A
Right, based on whoever's interacting.
Speaker A
In this case, then it's going to call the benefits enrollments and the public workers to get worker information.
Speaker A
As well as the benefits information.
Speaker A
The worker information is, you know, get the worker by the person number.
Speaker A
So get user session, then get the worker.
Speaker A
So this gives us, you know, things like address, or things like enrollment IDs.
Speaker A
And all the details of the person or the worker.
Speaker A
There's this document tool that has all the documents for our benefits.
Speaker A
So if I ask a question about my benefit.
Speaker A
It'll do a lookup and it will then, you know, provide me an answer.
Speaker A
And then the one that we have here is the claims tool.
Speaker A
Which, you know, it's the one that I just showed.
Speaker A
This is just a reference to that tool.
Speaker A
So yeah, let's run it and see it in action.
Speaker A
Just by typing hi, you will see it'll give me my person number, person ID, name.
Speaker A
Email and then some other details.
Speaker A
Right, and this happens because in the prompt, we've, we've tell the agent to.
Speaker A
You know, get all the details, whatever someone starts chatting.
Speaker A
And we can see the the string of calls it makes.
Speaker A
Right, first it, you know, takes the high in, uses an LLM to understand what the question is.
Speaker A
Or whatever the user inputted.
Speaker A
Next, it calls the get user session.
Speaker A
And when I click this, it'll show me, you know, the user details.
Speaker A
And particularly, we're getting the person number here.
Speaker A
But basically, it's internally just calling the business object to get the details of the user.
Speaker A
Next, it sends that to the LLM.
Speaker A
That, you know, this is the prompt.
Speaker A
This is what I got as a part of the tool response from get user session.
Speaker A
What do we do next, right?
Speaker A
So then it, you know, has a series of calls and LLM requests.
Speaker A
Which it uses to also create the structure from that JSON.
Speaker A
And understand whatever we've typed there.
Speaker A
Okay, next, I asked view current enrollments.
Speaker A
And it called the business object to get a list of enrollments that I have.
Speaker A
And it, you know, displayed in this sort of format.
Speaker A
Right, so up to this point, I asked hi and view my current enrollments.
Speaker A
And those were all internal calls to the tools internally.
Speaker A
No external calls made.
Speaker A
But when I ask the agent to get me a list of claims I made last year.
Speaker A
Using the instructions I gave it, it understands I need to call the OIC MCP tool.
Speaker A
And it returned the list of claims.
Speaker A
And in the details here, we can see if I scroll down to the point where it's making the call.
Speaker A
That it actually made the call to the get claims for employee API integration via the MCP tool that we created.
Speaker A
And if I come back to the integration.
Speaker A
Then I'm able to see what it sent to my integration was the employee ID.
Speaker A
And the response that I got back.
Speaker A
This is what we see in the Fusion AI agent.
Speaker A
So yeah, that was just a sample to show you how you can use Oracle integration as the middleware.
Speaker A
To broker communication between an AI agent and some sort of private resource.
Speaker A
With Oracle integration, you unlock endless possibilities.
Speaker A
You can, you know, connect to databases, you know, queues, topics, file servers.
Speaker A
Any sort of file-based integrations as well.
Speaker A
And then if there are, you know, systems where there aren't APIs, you can use the RPA tool to connect.
Speaker A
Pull information, push information.
Speaker A
All sorts of things.
Speaker A
You can kick off, you know, approval flows, human interaction flows, assign tasks.
Speaker A
Using the process automation, and then also the decision service to use complex decision trees and decision tables within your AI agent flows.
Speaker A
So yeah, thank you for watching.
Speaker A
Hope this was helpful.
Topics:Oracle IntegrationFusion AI Agent StudioMCP toolOracle CloudAI agentsLarge Language ModelFusion CloudOracle Integration CloudBusiness Object ToolOn-premises database integration
