Get Started with AI Agent Studio

You can start with a preconfigured agent template or create your own agent team.

When using a preconfigured agent team template, the artifacts such as agents, tools, and topics, aren't directly editable. To change these artifacts, you create a copy of the artifact and add the copy to the agent team.

Note: Agents created directly within an agent team are directly editable within that team.

Here's a broad outline of the tasks involved in creating an agent team.

Task	Details
Define tools	To effectively define the tools required by an agent, you need to first identify the types of questions users might ask and then decide which tools the agent needs to answer those questions accurately. These are the available tools: Oracle Fusion Cloud Applications business object: Business Object tools allow AI agents to retrieve, update, create, or delete business object records within Fusion Applications. Using this tool, agents can securely access information or call specific functions in the application. You can control what data the agents can access, by selecting the business objects and fields to use or ignore. AI agents adhere to the native security and role-based access controls of Fusion Applications, ensuring protection and privacy for your enterprise data. Document tool: You can upload specific documents to be used by the AI agent, and provide natural language instructions on how the agent should use these documents. The agent can then search for information in the documents to provide a more exact answer to a user's question. Email: This tool can access the email client to send emails that include summaries of interactions or details pulled from a knowledge store. Deep link: A deep link will send a user directly to the part of Fusion Applications where they can update underlying information. For example, if the user moved and wants to update their home address in the HR system, a deep link can quickly route the user to the page for making that update. External REST: You can connect to internal and external SaaS applications or public APIs using External REST tool. To connect to internal and third-party services, add the authorization parameters. For example, you can retrieve the real-time weather information for a specific location by creating an External REST tool that connects to the appropriate external API. MCP: You can use the Model Context Protocol (MCP) tool and securely connect to external MCP servers, without building additional REST wrappers or plugin logic.
Define topics	Topics define the focus of the agent to a specific area of expertise. They are prompts that you can reuse across multiple systems or summarization prompts. For example, within a Benefits Administrator agent, we might define topics such as health policy coverage, vision policy coverage, and benefits enrollment. Use Topics to efficiently streamline your interactions with the agent. Specify the instructions that help the agent decide which tools to use. Enable the agent to better understand user intent by letting it identify and select the relevant topic based on the user's question. Give each topic a clear, specific name, and include natural language instructions to ensure it's used correctly. You can reuse topics across agents.
Define credentials	To enable access to more services, you need to provide the necessary connection credentials. You can add credentials for this artifact: Custom LLM: In addition to the LLMs provided by Oracle, you can use other LLMs you've access to. You can add credentials for your LLM, and select it while creating your agent, node, or agent team.
Build new agents	Define the capabilities and scope of the agent. Agent name Product area the agent will work in Natural language instructions to allow the agentic flow, or other agents, to understand the capabilities of this particular agent Tools and topics the agent will need to use. In addition to the tools and topics you created, you can also add the predefined ones. In the Tools and Topics tabs, the predefined tools and topics are indicated by the icon.
Add a user (human) in the loop	If required, add an approval step for some actions that your AI agent will perform. The review step can be added at any point in the process for oversight and control over key actions, such as sending an email or updating a record.
Build the agent team	Create an agent team and add agents and other artifacts to it. Types of agent teams are: Supervisor: A supervisor agent manages other agents and artifacts in the agentic flow. Workflow: An agentic flow that does the tasks in a predetermined order. Agents and artifacts are added as nodes in the workflow, and each node performs a defined function, for example, extracting data, calling a business object, running an LLM, or sending an email. The node then passes its output to the next step. Note: You can also add predefined agents to your agent team. In the Agents tab, the predefined agents are indicated by the icon.
Test the agents	Make sure to test the agent before deploying to production. Ask a test query, and determine the accuracy and relevance of the agent's response. You can also see the instructions the agent is following, and the actions the agent has taken to arrive at the response.
Deploy the agent team	After defining and testing your agent team, you can deploy it directly from AI Agent Studio. Embed the agent conversation chat experience into any website or application. Trigger the agent from an external resource using Webhooks, or seamlessly embed the chat experience into HTML and React web pages.

RAG Agents to AI Agent Studio

If you've previously created any RAG agents in Fusion Applications, we recommend replacing your existing agent with a new one created in AI Agent Studio. For details, see Migrate Document Tools of RAG Agents.