Oracle AI Agent Studio is evolving rapidly, and with that has come a lot of confusion around terminology, architecture, and functionality. This crash course breaks down the current structure of Oracle AI Agent Studio, the different components within it, and some important things organizations should understand before getting started.
Overall Architecture of Oracle AI Agent Studio
Supervisor Agent Teams > Agents > Tools
/
Agentic Apps > Agent Teams
\
Workflow Agent Teams > Nodes + (Agents > Tools)
Chart Legend
Items to the left of the “>” are comprised of items on the right. Lines “/” or “\” indicate subtypes.
Agent Teams
As the name states, these are teams of agents. There are two types of Agent Teams: Workflow Agent Team and Supervisor Agent Team.
Supervisor agent teams are the original structure and seem to be gradually getting phased out in favor of workflow agent teams. Supervisor agent teams can only be comprised of agents, and there is a main agent that determines routing to subagents based on the user’s prompt. The user has limited control over that routing outside of adding prompt guidelines.
Workflow agent teams can be made of both agents and nodes. Nodes allow developers to have more control over routing once the user prompts the agent. Workflow agent teams also support additional functionality such as:
- Webhook triggers
- Chat triggers
- Scheduled triggers
- Error handling
Nodes
Nodes are only available within workflow agent teams and act as the building blocks of workflow agent teams.
5 Different Node Types
- AI
- Communication
- Data
- Logic
- Workflow Control
AI Nodes
LLM
Basic LLM prompt with text input. Returns response in either text or JSON format.
Agent
Calls an agent.
Workflow
Calls another workflow agent team.
RAG Document Tool
Allows semantic search on documents.
Communication Nodes
Send Email
Allows workflow agent teams to send emails as part of an automated process. This can be used to deliver notifications, summaries, approvals, or other generated content to users. Combined with workflow triggers and approval nodes, email functionality can support use cases such as status updates, exception notifications, approval requests, and follow-up communications within a broader agent workflow.
Data Nodes
Document Processor
Likely using Oracle Document Understanding/OCR functionality. Takes in documents and parses them into JSON format based on configured criteria.
Business Object
Functions similarly to Fusion business objects.
External REST
Calls external REST APIs.
Tool
Workflow agent team tools can currently only be:
- Deep Link
- Chat Attachments Reader
- Intent Change Indicator
- User Session
Tools created inside workflow agent teams appear to function differently than tools created in the standalone Tools section of AI Agent Studio. Based on current functionality, tools created in the Tools section can only be used by agents.
Vector DB Reader
Searches vector databases using semantic meaning.
Current limits:
- 5 results per read
- 15 total retrievals per session
- 10 semantic searches + 5 text-only searches
Vector DB Writer
Writes semantic meaning into a vector database for future semantic retrieval.
Current limits:
- 50KB max document size
- No more than 500 objects per agent team
Logic Nodes
Code
Basic JavaScript execution.
Set Variables
Sets variables from node outputs. Variables must first be declared in the workflow agent team settings.
Workflow Control Nodes
Human Approval
Adds prompt in chat for human confirmation. Can be a simple yes/no approval or updated user input.
If Condition
Routes execution based on true/false conditions.
For Loop
Loops for a specified number of iterations.
While Loop
Loops while conditions remain true.
Switch
Changes routing based on cases.
Run in Parallel
Executes processes simultaneously.
Wait
Pauses execution.
Return
Returns response to the user.
Agents
Agents can only have tools attached to them. They are essentially LLMs with attached tools.
Available Tool Types
- Business Object
- Connector (external web searches)
- Deep Link
- Document
- External REST
- MCP
Agent Configuration
When configuring an agent, developers can edit:
- Agent Persona
- Prompt
The Agent Persona is intended to define tone and role-based behavior. Agents can also have Topics attached to them. Topics act as reusable instructions for agents.
Agentic Applications
In its current state, an agentic application is effectively a dashboard containing several agent teams.
Oracle has stated that workflow agent teams should act as the primary building blocks for agentic apps. Users can prompt for information, and the agent teams work together in some fashion to deliver responses through visuals or text outputs.
Agentic Application Capabilities
- Use pre-seeded functionality
- Use prompt-driven interactions
- Display visual or text responses
Testing and debugging functionality still appears to be in progress. While individual agent teams can be tested independently, troubleshooting issues at the full application level is still limited.
Terminology Clarifications for AI Agent Studio
LLM Node vs. Agent Team vs. Agent vs. Agentic Apps vs. AI
One challenge when learning AI Agent Studio is that some terms sound similar, and in some cases the same term may be used to refer to different things depending on the context. Understanding the distinctions between these components can help make the overall architecture easier to follow. So, to reiterate:
- An LLM Node is a workflow component that sends prompts to a large language model and returns a response in either text or JSON format
- An Agent is essentially an LLM with tools attached to it
- An Agent Team is a collection of agents working together. Depending on the type, agent teams may also include workflow nodes and routing logic
- An Agentic Application is a higher-level experience that brings together one or more agent teams to deliver information, actions, visuals, or other outputs to users
- AI (Artificial Intelligence) is a broad industry term that refers to systems, software, algorithms, or tools designed to emulate aspects of human intelligence, such as reasoning, decision-making, language processing, learning, or problem-solving
Understanding how these components relate to one another can make it easier to navigate AI Agent Studio and determine which functionality is best suited for a given use case.
Vector DB Reader/Writer vs. Document RAG Tool vs. Document Processor Tool
There is still overlap in how these are described and when each should be used.
Agent Tools vs. Workflow Agent Team Tools
The functionality appears different depending on where the tools are configured.
When to Use Supervisor Agents vs. Workflow Agents
From a developer perspective, workflow agents currently provide significantly more control and flexibility. Supervisor agents may still offer faster setup for simpler use cases, but workflow agent teams appear to be where Oracle is focusing future development.
Things to Note for Oracle Fusion Cloud Users
Many organizations are currently in the basic tier for Oracle AI Agent Studio.
Included Functionality in the Basic Tier
According to Oracle, this tier allows:
- Instantiating Oracle agent templates and deploying them without custom changes
- Minor contextual changes such as uploading documents or adding custom fields
- Editing prompts without changing the overall purpose of the agent
- Adding Topics for additional guardrails or tone adjustments
- Adding email tools
- Adding Deep Links to Oracle pages
- Expanding business objects with additional fields or columns
- Uploading PDFs, policies, or FAQs into provided Document Tools
- Development and testing using Oracle-hosted LLMs
This effectively means that if organizations encounter issues with Oracle-provided template agents and do not have the custom AI Agent subscription, the available changes are mostly limited to smaller prompt and configuration adjustments.
Oracle AI Agent Studio is still evolving rapidly though, and both provided and custom functionality appear to be changing frequently with new updates.
What’s Next
Oracle AI Agent Studio is still maturing, but workflow agent teams, orchestration capabilities, and agentic applications are clearly becoming an important part of Oracle’s broader AI strategy. Understanding the current architecture, limitations, and terminology is an important first step for organizations evaluating how these capabilities may fit into their Oracle environment.
For more Oracle AI insights, demos, and educational content, subscribe to the Elire YouTube channel and follow along with our latest AI-focused videos and discussions. You can also subscribe to our Oracle Cloud newsletter to stay up to date on Oracle Fusion Cloud, AI innovation, and product updates.
If your organization is evaluating Oracle AI Agent Studio, workflow automation, or broader Oracle Fusion Cloud initiatives, explore Elire’s Oracle Fusion Cloud services to learn how our team can help support your strategy and implementation goals.
