Introduction
Oracle AI Agent Studio enables organizations to rapidly build AI-powered assistants embedded within Oracle Cloud Applications. In this walkthrough, I demonstrate how to create, configure, test, and publish a custom procurement-focused AI agent using an out-of-the-box template.
Procurement users often spend time manually checking requisition status across systems. This agent simplifies that by enabling natural language queries tied directly to transactional data.
What You Will Build
- A custom AI Agent for Purchase Requisition status
- Integration with business objects and document retrieval (RAG)
- Defined topics to control agent scope
- A tested and published AI Agent Team
Access Oracle AI Agent Studio
Navigate to Tools → AI Agent Studio within Oracle Cloud Applications. This provides access to prebuilt agent templates that accelerate development.
Create a Custom Agent from Template
Filter templates by Procurement and select a relevant out-of-the-box agent. Use the 'Copy Template' feature to create a custom version. Apply a naming convention (e.g., adding a suffix) to distinguish your agent.
Key Components of an AI Agent
Tools
Defines the additional utilities an agent can use to accomplish a task. One or more tools are assigned to agents, and they're reusable among agents.
List of Tools:
• Calculator tool
• Email tool
• Business object tool: Use Business Object tools when you need structured transactional data (e.g., PR status).
• User query tool
• Document retrieval tool for retrieval-augmented generation (RAG) : Use Document Retrieval (RAG) when answers depend on unstructured knowledge (e.g., policies or documents).
Topics
Defines the areas of expertise through instructions that set the boundaries and constraints for agent conversations and abilities.
Without well-defined topics, the agent may respond outside procurement scope, reducing accuracy.
Configure Agent Details
Review agent configuration including prompts, LLM provider selection, and tool assignments. Save the agent in draft status for further testing.
Configure Agent Team
In Agent Team settings, configure LLM providers, define security roles, and set starter questions to guide user interaction.
Currently, three LLM options are available: GPT-5 mini, GPT-4.1 mini, and GPT-OSS-120B. Rather than defaulting to the latest model, selection should be based on the use case.
GPT-5 mini: strong balance of reasoning capability and efficiency; suitable for most enterprise workflows
GPT-4.1 mini: faster and more cost-efficient; works well for simpler, high-volume interactions
GPT-OSS-120B: more flexible and customizable; useful when control or specific tuning is required
In practice, model choice is a tradeoff between capability, latency, token usage, and cost. For most scenarios, the “best” model is not the most advanced one, but the one that meets requirements with the lowest operational overhead.
Test Using Debug Mode
Click on the Triangle to run in debug mode.
Use the Debug feature to validate agent responses. Ensure queries return accurate procurement data and remain within defined topic boundaries.
Example: “What is the status of my purchase requisition?”
The agent retrieves real-time data using the business object tool and returns the current approval status.
Publish the Agent
Once validation is complete, publish the agent team to make it available to end users.
Key Takeaways
- Templates significantly reduce setup time
- Tools extend agent capabilities through integrations
- Topics are critical for controlling agent scope
- Debugging ensures reliability before deployment
Top comments (0)