ORACLE CONFIDENTIAL. For authorized use only. Do not distribute to third parties.

Pre-General Availability: 2025-12-16

Create Custom Flows with Agent Builder

Use Agent Builder to leverage enterprise data for agentic use cases and to build custom flows.

From the left sidebar, navigate to Agent Builder or My Custom Flows to create custom workflows.

See Agent Builder Components to learn about the components available in Agent Builder.

Example Use Case 1 - Bug Assistant

Create an agent that offers a natural language interface for Bug DB. The intelligent agent integrates with BugDB APIs to collect relevant data and deliver accurate, timely responses to user queries related to bug triage, including prioritization, status updates, assignments, and other bug details.

  1. Start with a fresh canvas.

  2. Drag and drop Agent component to the canvas.

  3. Select an LLM from the dropdown and fill Custom Instructions:

    You're a helpful assistant in a software team.
Help users triage software bugs using natural language.
You understand how to interact with the bug triage system.
Your task:
- User wants to triage software bugs. Parse their natural language request and select the appropriate tool for it.
- User expects a response that confirms an action or provides required information.
Important:
- Be helpful and concise in your messages
- Do not tell the user any details not mentioned in the tool response, let's be factual.

    You can also choose to receive custom instructions from Chat Input or Prompt by connecting either of the components as inputs to Agent’s Custom Instructions.

  4. Drag and drop Bug Tools component to the canvas and connect it to Tools input of Agent component.

  5. Drag and drop Chat Output to the canvas and connect Agent to Chat Output.

  6. Edit Workflow Name and Description.

  7. Publish the workflow.

  8. Click Playground to start interacting with the assistant.

    Sample Questions:

    • What is the subject of the bug 38086462?

    • Who filed the bug 38086462?

    • When was the bug 38086462 last updated?

Example Use Case 2 - Diagnostics and Documentation Assistant

Create an agent that integrates with two MCP servers - AHF Insights and DBSA, to collect oracle diagnostic information, analyze system health, and refer custom documentation, facilitating in-depth agentic exploration and troubleshooting support.

  1. Start with a fresh canvas.

  2. Drag and drop Agent node to the canvas.

  3. Select an LLM from the dropdown and fill Custom Instructions:

    You're a helpful assistant in a software team.
Help users triage software bugs using natural language.
You understand how to interact with the bug triage system to check bug status, fetch attached files, parser status, insights reports associated with the bug, events present in the bug and issue recommendations.
Your task:
- User wants to triage software bugs. Parse their natural language request and select the appropriate tool for it.
- User expects a response that confirms an action or provides required information.
Important:
- Be helpful and concise in your messages
- Do not tell the user any details not mentioned in the tool response, let's be factual.
- Use documentation tool to provide answers for questions around product knowledge.
- Format all the responses in markdown.

    You can also choose to receive custom instructions from Chat Input or Prompt by connecting either of the components as inputs to Agent’s Custom Instructions.

  4. Drag and drop two MCP Server component to the canvas and enter the MCP Server URL to get tools to use with the agent.

  5. Connect the MCP Server to the Tools connector of Agent component.

  6. Drag and drop Chat Output to the canvas and connect Agent to Chat Output.

  7. Edit Workflow Name and Description.

  8. Publish the workflow.

  9. Click Playground to start interacting with the assistant.

    Sample Questions:

    • Show cluster summary for the bug 38086462.

    • What are the top features of Grid Infrastructure Version 19.26.0.0?

Example Use Case 3 - Email Assistant

Create a custom flow to create and send product pitch emails.

  1. Start with a fresh canvas.

  2. Drag and drop Agent component to the canvas.

  3. Select an LLM from the dropdown and fill Custom Instructions:

    Create a concise, persuasive product pitch that clearly explains what the product is, its unique value, 
and the main benefits for its target audience. End with a strong call to action. 
Keep it under 150 words and use a professional, engaging tone. 
The email should be an HTML body with HTML tags for formatting headings and texts.

    You can also choose to receive custom instructions from Chat Input or Prompt by connecting either of the components as inputs to Agent’s Custom Instructions.

  4. Drag and drop Chat Input to the canvas and connect it to Agent’s Prompt input.

  5. Drag and drop Email Output to the canvas and enter recipient’s email id and email subject.

    Note: You must configure SMTP before you use the Email component. See Configure SMTP.

  6. Connect Agent’s output to Email Output.

  7. Drag and drop Chat Output to the canvas and connect Email Output to Chat Output.

  8. Edit Workflow Name and Description.

  9. Publish the workflow.

  10. Click Playground and provide details about your product in the chat. The agent will then generate a compelling product pitch and send emails to the recipients.

Example Use Case 4 - Refund Assistant

Create a Refund Assistant that uses hierarchical manager-worker agent orchestration to process customer refund requests. The manager agent receives the initial refund request and delegates tasks to specialist worker agents or sub-agents equipped with client and server tools to make a final decision.

  1. Start with a fresh canvas.

  2. Drag and drop an Agent component to act as the Refund Manager.

  3. Add three Agent components to act as worker agents - Order Verifier Specialist, Policy Checker Agent, and Resolution Specialist Agent.

  4. Connect these sub-agents to the manager.

  5. Configure the Order Verifier Specialist Agent with an MCP Server tool to fetch order details given an order ID. Select a LLM for the agent and fill Custom Instructions:

    You are an Order Verifier specialist. 
Given a refund request with order ID and issue, retrieve and confirm order details like items, delivery time, and status. 
Use tools to query data if needed. 
Respond with verified facts only.

  6. Configure the Policy Checker Agent with an MCP Server tool to fetch refund policy. Select a LLM for the agent and fill Custom Instructions:

    You are a Policy Checker for refunds. 
You should fetch the policy and check if the user is elegible for refund or not based on the Order Details.

  7. Configure the Resolution Specialist Agent with an MCP Server tool to decide on refund amount. Select a LLM for the agent and fill Custom Instructions:

    You are a Resolution Specialist. 
Given policy eligibility and order details, decide on the refund amount (full/partial/none) and generate a customer response. 
Prioritize customer satisfaction while following policies. 
Include next steps like refund processing time.

  8. Define the Refund Manager Agent’s Custom Instructions to orchestrate these sub-agents based on user queries.

    You are the Refund Manager. 
For each refund request, delegate tasks as follows: 
To the Order Verifier for order details, to the Policy Checker for refund eligibility, and to the Resolution Specialist for the final decision. 
Synthesize their responses and provide the overall outcome. 
If needed, ask for more info.

  9. Connect a Chat Input component to the Refund Manager and connect the output to a Chat Output component.

  10. Publish the workflow.

  11. Click Playground to interact with the assistant. The Refund Manager Agent will delegate tasks per the instructions and provides a comprehensive customer support experience.

    Sample User Query: “I want to request a refund for Order ID 126. It’s not delivered yet.”