Skip to main content

OIC AI Agents: Understand Project AI Agents Through Familiar Architecture

Calculating read time…

A practical orientation for experienced Oracle Integration developers beginning their journey into agent development. Connect familiar integration architecture to the new concepts of agents, thinking patterns, models, tools, and project-level observation.

Module boundary: This is a workspace-exploration and design exercise. We will inspect an available training project, map a purchase-order question to its proposed components, and define a read-only Procurement Assistant. We will not create integrations, configure an LLM, activate an agent, or expose an API in this post.

1. What You Will Build

Consider a common procurement-support question:

Sample user request

"What is the current status of purchase order PO-450017? Please give me the supplier, total amount, and latest available status."

A conventional integration can retrieve a purchase order through a predefined API. An agent adds a different capability: it can interpret a natural-language request, select an eligible registered tool, use the information returned, and formulate an answer appropriate to the question.

For this module, our proposed solution is a Procurement Read-Only Assistant.

Design element Proposed outcome
AgentProcurement Read-Only Assistant.
Business purposeAnswer supported questions about existing purchase orders.
Proposed toolA future registered OIC integration that retrieves an approved set of purchase-order fields.
Target systemThe organization's authorized procurement source of record.
Allowed behaviorRead, summarize, identify missing information, and report uncertainty.
Excluded behaviorCreating, editing, approving, cancelling, or closing purchase orders.

Here is an illustrative response:

Illustrative response — fictional data, not an actual ERP result

Purchase order PO-450017 is currently Approved. The supplier is Atlas Office Supplies, and the total order amount is USD 1,840.00. In a real implementation, the status and amount must come from the authorized procurement system's returned data.

This module does not produce a working agent. Its deliverables are a workspace inventory, an architecture sketch, and an Agent Charter that will guide later implementation.

2. In This Post

3. Before You Begin

Required access and assets

Prepare the following before beginning the workspace exercise:

  • An authenticated OIC session.
  • Permission to view the relevant project and its design-time assets.
  • An existing authorized training project for the navigation exercise; if none is available, complete only the offline design work.
  • The actual instance release and relevant feature availability, confirmed with your OIC administrator if uncertain.
  • A place to record the workspace components and Agent Charter.

For shared projects, ask the project owner or administrator to confirm your view/monitor permissions before treating a missing page or empty list as a feature limitation. Access to the project does not automatically establish access to every business system.

No LLM API key, model connection, procurement connection, integration endpoint, or agent configuration is required. Do not enter credentials or real business data into sample prompts or learning notes.

Asset-readiness checklist

Asset Disposition
OIC instance and training projectReuse an existing authorized environment.
Existing agents, tools, agent patterns, and prompt templatesInspect and record what exists; do not modify it.
Procurement integration and live purchase ordersNot required. Use the fictional design scenario.
Architecture sketch and Agent CharterCreate these as learning deliverables outside OIC.

If no training project is available, do not create a new project merely to complete this lesson. Complete the design exercises, mark workspace inspection as blocked, and resolve the access prerequisite before beginning implementation.

4. Essential Agent Concepts

The business problem: When does an agent add value?

Enterprise requests do not always arrive as structured API payloads. A procurement user might ask about a purchase order, explain a supplier issue, or request several related facts in a single message. The system must interpret the question before retrieving the required business information.

An agent can be useful when a request needs natural-language interpretation, selection among authorized capabilities, consideration of returned information, and an answer adapted to the user's question.

Business need Suitable starting point
Transform a known payload using fixed rules.A conventional OIC integration.
Explain supplied text or generate a short summary.An LLM call may be sufficient.
Interpret a question, select an eligible information-retrieval tool, and explain the returned information.An agent working with governed tools.
Execute a regulated approval or financial transaction.A controlled workflow or integration with enforceable authorization and validation; an agent may assist where appropriate.

The decision is not whether a process can be described as an agent. The question is whether model-guided interpretation and capability selection add enough business value to justify the additional testing and operational controls.

Agent vs LLM call vs integration

These components solve different problems. They can work together, but they are not interchangeable.

Component Main responsibility Procurement example
LLM callInterpret or generate language from supplied context.Explain a supplied definition of "Pending Approval."
OIC integrationPerform configured operations, transformations, routing, and system calls.Call an authorized purchase-order retrieval API and normalize its response.
OIC Project AI AgentCoordinate model-guided interpretation and use of registered tools within its configured scope.Interpret the question, select a suitable retrieval tool, use its result, and formulate an answer.

Analogy: The LLM supplies language and reasoning capability, the integration provides a configured execution path, and the agent coordinates when to use its available capabilities. This analogy describes their roles; it does not mean the model can execute arbitrary code or access enterprise data by itself.

An integration that calls an LLM does not automatically become an agent. Likewise, creating an agent does not automatically give it access to Oracle Fusion Cloud Applications, databases, or external services. Actual access depends on the tools, connections, endpoints, and controls provided to it.

The main components

Oracle documents distinct areas for agents, tools, agent patterns, and prompt templates in the OIC Projects workspace. The following map connects those design-time components to the model and business system.

Component What it contributes Our design
AgentThe configured assistant, including its role, guidelines, and eligible tools.Procurement Read-Only Assistant.
Agent patternThe chosen thinking-pattern recipe and configuration used to coordinate model reasoning and tool actions (for example, ReAct or Plan and Execute, where available).A suitable available pattern recipe will be evaluated in a later implementation module; none is selected here.
Model and LLM connectionThe LLM connection and its provider/model settings associated with the agent pattern; not a separate source of procurement data.Not selected or configured here.
Agentic AI toolA registered capability that the agent can select for a defined task.A proposed read-only purchase-order retrieval integration.
Prompt templateA reusable prompt structure that can accept supplied dynamic data.Optional for this basic design; not created here.
Target systemThe authoritative provider of business records and their current state.The authorized procurement source of record.

A knowledge base is another project capability. It can support searches across prepared documents, but it is not automatically a live query against a transactional procurement system. For current purchase-order status, a suitable read-only transactional lookup is the proposed source of truth.

See Oracle's Plan Your AI Agent guidance for identifying the agent's purpose and required tools.

Role, guidelines, pattern, and tool metadata

These elements can contribute to the context supplied to the model, but they have different responsibilities.

Element Question it answers
Agent roleWho is this assistant intended to be, and what is its business purpose?
Agent guidelinesWhat should this assistant do, avoid, or disclose?
Pattern guidelinesHow should the selected pattern coordinate reasoning and actions?
Tool description and guidelinesWhen is a particular capability suitable, and what usage constraints should inform selection?
Tool parametersWhich inputs can be supplied, and which values or options apply?
Prompt templateHow can a reusable prompt structure receive dynamic input?
Business dataWhat did the actual source return for this request?

Oracle's tool-registration documentation distinguishes the tool name, identifier, description, guidelines, and input-parameter configuration. The identifier is sent as tool metadata; the description and guidelines help the model understand appropriate use. Parameter visibility determines which parameters are sent to the LLM. These settings influence tool selection, but they do not replace authorization or backend validation.

For the exact registration contract, see Register an Integration as an Agentic AI Tool.

Oracle product boundaries

Oracle offers several agent-related capabilities. Similar terminology does not mean they share the same designer, runtime, resources, deployment model, or integration contracts.

Capability Primary context Boundary to remember
OIC Project AI AgentsAgent development within Oracle Integration 3 projects.Uses project assets and the OIC agent, tool, and pattern capabilities.
Oracle AI Agent Studio for Fusion Cloud ApplicationsCreating, configuring, validating, and deploying agents and workflows integrated with Fusion Cloud Applications.It is a Fusion Applications design environment, not the OIC Projects workspace.
OCI Generative AI agent capabilitiesBuilding agentic applications using OCI Generative AI APIs or deploying hosted agentic applications.Uses its own APIs, resources, and deployment options.
OIC MCP GatewayGoverned access for MCP clients to approved capabilities exposed through managed MCP servers.MCP exposure is a distinct access path; do not confuse it with an agent's registered project tools.

Oracle describes Fusion AI Agent Studio as integrated with Fusion Cloud Applications. OCI Generative AI separately documents API-based agent workflows and hosted agentic applications. These are different product contexts from the Project AI Agents workspace covered here.

Primary references: Overview of AI Agent Studio, Enterprise AI Agents in OCI Generative AI, and the OIC 26.10 release notes for MCP Gateway.

Boundary check: Finding a similarly named feature in Fusion or OCI documentation does not establish that the feature, API, menu, configuration, or runtime procedure exists in OIC Project AI Agents. Verify the exact product documentation before implementing it.

The Project workspace

The Projects workspace brings project resources into Design, Deploy, and Observe tabs. In Design, the AI Agents page contains separate Agents, Tools, Agent patterns, and Prompt templates sections. Availability in your instance and your ability to inspect items depend on the environment and assigned access.

  • Design: Inspect the AI Agents page and related project assets.
  • Deploy: Locate the project deployment area. No deployment is required in this exercise.
  • Observe: Locate agent-instance and integration-monitoring areas. Without a test run, this lesson will not generate runtime evidence.

The key difference for an experienced integration developer is that a project can contain agent-specific resources alongside familiar integration assets. Those agent resources need to be understood as separate design elements.

The Agent Charter

Before creating an agent, define what it is permitted to accomplish. Here, the Agent Charter is an engineering design document—not an Oracle configuration object or an enforcement mechanism.

A useful charter specifies intended users, business purpose, supported operations, forbidden actions, evidence requirements, and handling of uncertainty. We will complete it during the practical exercise.

5. Practical Prototype — Step-by-Step

This prototype has two outputs: an inventory of the available workspace and a design for the read-only Procurement Assistant. It does not create or change OIC assets.

Phase 1 — Open the project and establish its boundaries

Step 1 — Enter the Projects workspace

Starting point: An authenticated Oracle Integration 3 session with permission to inspect a training project.

Navigation: OIC navigation pane → Projects → select the available training project.

Actions:

  1. Open Projects from the OIC navigation pane.
  2. Find and select the project you are authorized to inspect.
  3. Confirm that you are viewing the intended project and environment.
  4. Locate the Design, Deploy, and Observe tabs.
  5. Record the project name and available release information.

Configuration: None. Do not create a project or modify its details.

Expected result: The selected project's workspace opens, with project-level navigation available according to your permissions.

Checkpoint: Record whether the project is accessible and whether you can locate all three tabs. If a project or tab is missing, investigate access and environment prerequisites instead of assuming the feature does not exist.

Documentation: Tour of the Projects Workspace.

Step 2 — Locate the AI Agents page

Starting point: The selected project's Design tab.

Navigation: Projects → select project → Design → AI Agents in the project's vertical navigation.

Actions:

  1. Open the project's Design area.
  2. Locate the AI Agents page.
  3. Open the page without selecting Add, Create, Activate, or another modifying action.
  4. Inspect the available sections and record existing asset names and visible statuses, if shown.

What to record: Whether the Agents, Tools, Agent patterns, and Prompt templates sections are present and which assets, if any, are visible.

Expected result: The project-level AI Agents page opens, presenting the agent-related resources available for inspection.

Checkpoint: Keep existing assets separate from the proposed Procurement Assistant. Do not assume an existing item is suitable for this scenario or create anything to fill an empty inventory.

Oracle documents these sections in its Projects workspace tour.

Step 3 — Inspect the surrounding project areas

Starting point: The project's Design tab.

Navigation: Use the project's vertical navigation to inspect Integrations and Knowledge Base, then return to AI Agents.

Actions:

  1. Inspect the Integrations area to understand where conventional project integrations are listed.
  2. Locate Connections and note that integrations may depend on connection assets.
  3. Locate Knowledge Base and note whether existing knowledge resources are present.
  4. Return to AI Agents and limit your inventory to resources relevant to the proposed assistant.

Expected result: You can distinguish familiar integration assets from agent-specific components and recognize Knowledge Base as a separate project capability.

Checkpoint: Do not edit, create, activate, or delete any asset.

Phase 2 — Inspect observation areas

Step 4 — Locate agent and integration observations

Starting point: The selected project's workspace.

Navigation: Project → Observe → AI Agents. Separately inspect the integration-monitoring area when needed.

Actions:

  1. Open Observe.
  2. Locate AI Agents and its Agent instances tab, if available.
  3. Identify the documented observation fields: agent name, conversation ID, status, user prompt, and duration.
  4. Switch to the Observe integration page and locate its Integrations and Instances tabs; these provide integration-level summary and execution details when runs exist.
  5. Record what monitoring capabilities are visible without starting a run or changing an instance.

Expected result: You can explain where agent-level observations live and where integration execution evidence can be inspected.

Important: This is an inspection exercise. If the project has not executed an agent, an empty instance list is not evidence of an agent failure. It means no observed run is available for this exercise.

See Oracle's workspace tour and Run and Test the AI Agent documentation.

Phase 3 — Map the purchase-order question to components

Step 5 — Define the proposed request and data contract

Starting point: The fictional Procurement Assistant scenario defined in Section 1.

Actions:

  1. Write down the business question the assistant must answer.
  2. Identify the minimum input needed to find the purchase order.
  3. List only the fields required in the answer.
  4. Identify the authoritative source for the data.
  5. Specify how the assistant should respond if the source returns no match, missing fields, or an error.

Illustrative future tool input:

{
  "purchaseOrderNumber": "PO-450017"
}

Illustrative future tool response:

{
  "purchaseOrderNumber": "PO-450017",
  "supplierName": "Atlas Office Supplies",
  "status": "APPROVED",
  "currency": "USD",
  "totalAmount": 1840.00,
  "lastUpdated": "2026-10-20"
}

These are fictional, tutorial-defined example fields, not an Oracle-defined purchase-order schema or a tested API contract. The sample lastUpdated value is a record-update timestamp only; it does not prove when the business status last changed. Define the meaning and freshness of each field with the source owner before implementation.

Future implementation dependency: Oracle's documented registration process requires an eligible integration to be in a project and active, with a REST Adapter trigger as the integration's first connection, OAuth authentication, and a JSON payload. Oracle documents support for the REST Adapter's supported HTTP verbs. For this read-only scenario, an appropriately authorized GET retrieval operation is a sensible design choice. The eventual trigger contract must still be checked against the REST Adapter configuration, and the downstream service account and target API must be unable to mutate purchase orders. This dependency is recorded for a later module; do not build or activate that integration here.

Proposed field contract (design only):

FieldMeaning and handling
purchaseOrderNumber (request)Required business identifier; ask for it if absent. Validate form and entitlement in the backend.
supplierName, statusDisplay only values returned by the authoritative read service; do not infer missing fields.
totalAmount, currencyPresent together, preserving currency and precision defined by the source.
lastUpdatedOptional record timestamp; define timezone and semantics. Not necessarily a status-change date.

Failure cases to define: Record not found, requester not authorized, malformed identifier, source timeout, and incomplete authorized response must have distinguishable safe outcomes. This example does not prescribe HTTP status codes or the future ERP API response schema.

Expected result: A clearly defined question and a minimal input/output contract suitable for a future read-only retrieval tool.

Checkpoint: Every factual field in the future answer must have a defined source. The assistant must not manufacture a PO status or amount if the tool has not returned it.

Reference: Register an Integration as an Agentic AI Tool.

Step 6 — Draw the architecture

Starting point: The input/output contract from Step 5 and the workspace inventory from Phase 1.

Actions:

  1. Place the requester at the start of the request path.
  2. Place the OIC Project AI Agent after the request entry point.
  3. Show the configured agent pattern and LLM connection as the model-guided decision components.
  4. Show a proposed registered read-only integration tool.
  5. Place the procurement source system behind that integration.
  6. Draw the returned data path back through the tool to the agent and then to the requester.
  7. Mark the architecture as proposed, not deployed.

Proposed architecture sketch:

Authorized requester
        |
        | Natural-language question
        v
OIC Project AI Agent
  - Role and agent guidelines
  - Available tool metadata
        |
        v
Configured agent pattern
  + LLM connection and model
        |
        | Select an eligible retrieval tool
        v
Registered read-only agentic tool
        |
        | Authorized request
        v
Procurement source system
        |
        | Purchase-order data / not found / error
        v
Read-only integration response
        |
        v
Agent formulates a grounded answer
        |
        v
Requester

This is a conceptual request path, not a literal wire-level trace or an assertion about internal implementation details. The actual runtime flow must be verified after the integration, tool, pattern, and agent are configured.

Checkpoint: The diagram must contain an explicit business-data source. It must also show that the integration retrieves the record and that the final answer depends on returned information.

Phase 4 — Write the Agent Charter

Step 7 — Define users, responsibilities, and exclusions

Starting point: The architecture sketch and business request in Section 1.

Actions: Document the intended audience, supported requests, permitted capabilities, exclusions, evidence expectations, and failure handling.

Charter field Procurement Assistant definition
NameProcurement Read-Only Assistant.
Intended usersAuthorized procurement and accounts-payable support users.
MissionAnswer supported questions about existing purchase orders using authorized source data.
Allowed actionsRetrieve approved PO fields, summarize results, request a missing identifier, and disclose unavailable information.
Forbidden actionsCreate, edit, cancel, approve, release, close, or otherwise mutate purchase orders.
Data boundaryReturn only approved fields that the requester is authorized to access.
Evidence ruleUse returned source data for factual claims; never imply a record was retrieved when the tool was not called or its result is unavailable.
Failure handlingAsk for a missing identifier, explain a not-found result, or report a safe error without inventing a business status.

Expected result: A written definition of the agent's business boundary that a technical reviewer can inspect before implementation.

Checkpoint: The read-only requirement must also be enforced in the integration and target-system authorization design. Writing "do not update a PO" in agent instructions is not enough.

Finished design deliverable

At the end of this phase, you should have three artifacts: a workspace inventory, a proposed request-path diagram, and the written Agent Charter. These are design outputs—not deployed OIC resources.

6. Test and Prove the Outcome

Because this module creates and invokes no agent, the following checks are design and workspace checks—not runtime test results. The additional runtime scenarios define what must be tested after implementation.

Design checks for this module

Case Expected behavior Evidence to collect
Workspace accessOpen the intended project and locate its documented areas.Project name, release information, and observed navigation. Reader must verify; no execution evidence is supplied.
Component mappingIdentify the agent, pattern/model, future tool, and target data source.Workspace inventory and architecture sketch. Reader must verify; no execution evidence is supplied.
Read-only boundaryThe charter explicitly excludes purchase-order mutations.Completed charter and reviewer sign-off. Reader must verify; no execution evidence is supplied.
Source-of-truth pathThe planned answer depends on an authorized tool response.Data contract and architecture sketch. Reader must verify; no execution evidence is supplied.

Runtime cases for a subsequent module

Case Expected behavior Evidence to collect
Valid existing PORetrieve the matching record and answer from the returned fields.Tool input/output, source response, and agent-run evidence.
Missing PO numberAsk for the missing identifier instead of guessing.User prompt and observed agent behavior.
Unknown PO numberReport that no matching record was returned.Backend not-found result and final answer.
Request to cancel or approve a PODecline the unsupported action; do not expose write capability in the read-only design.Tool inventory, backend permissions, and execution evidence.
Unauthorized record accessEnforce access restrictions in the responsible identity, integration, or target-system layer.Identity context, authorization decision, and denied-operation evidence.
Model or tool failureDo not manufacture a successful result; provide a safe failure response.Agent activity and relevant integration-instance details.

All runtime scenarios are Not executed in this module. They become runnable only after the actual agent, model connection, tool integration, and required controls have been configured and tested.

7. Troubleshoot and Recover

For a workspace-exploration lesson, the common issues are access, feature visibility, or a misunderstanding about what an empty observation page means.

Symptom Targeted checks Recovery and retest
Projects or the intended project is unavailable.Confirm the instance, environment, project name, and assigned permissions.Request access or use an approved training project, then repeat Phase 1.
AI Agents is not visible.Confirm you are in the project Design area; inspect actual navigation and check permissions and feature availability.Compare the environment with current Oracle documentation and consult the administrator if unresolved.
A particular pattern or template is absent.Check the selected project, visible resources, and feature set for the instance.Record the observed state and verify prerequisites before later configuration.
Observe contains no agent instances.Confirm an earlier test actually ran and the correct project is selected.Record that runtime evidence is unavailable; do not invent an instance or claim a failure.
The agent is assumed to have PO access without a tool.Check whether a suitable registered tool and authorized data path have been designed and implemented.Update the architecture and tool contract before implementation.

Cleanup: No cleanup is required because this module creates no OIC assets. Keep the workspace inventory, architecture sketch, and Agent Charter for the next module. Leave existing resources and runtime state unchanged.

8. Corporate Implementation Notes

Identity and authorization

An agent's conversation and instructions do not automatically establish the identity or permissions of the person asking the question. A later implementation must define caller authentication, identity propagation where appropriate, and where authorization is enforced.

For the read-only Procurement Assistant, access to purchase-order information must be restricted by the responsible system or service. The future integration should expose only necessary read operations. If the target system supports user-specific entitlements, the architecture must explicitly preserve or enforce those entitlements as required.

Model instructions are not security controls

Role and guideline text can steer model behavior, but they should not be the sole control preventing unauthorized access or updates. A user can request an excluded action, and business data can contain misleading or instruction-like text. The integration and target system must continue to enforce their responsibilities regardless of what the model requests.

Limit information supplied to the agent

A tool intended to retrieve PO number, supplier, status, amount, and currency should not expose unrelated internal fields without a business justification. Oracle's tool-registration documentation includes parameter visibility and descriptions. The future implementation must review which inputs are presented to the model and which data can be returned to the requester.

Reliability and evidence

Treat model decisions, integration execution, and business effects as separate things to verify. An agent can produce a plausible sentence even if a tool failed or the source returned insufficient data. Runtime acceptance should compare the final answer with the actual tool response and corresponding integration evidence.

Ownership and lifecycle

Before deployment, establish ownership of the agent charter, tool contracts, model connection, target-system access, testing, operational monitoring, and future changes. Identify resource dependencies and follow the project's supported deployment process. This module does not create, activate, or deploy resources, so no deployment dependency is being asserted as satisfied.

9. Completion Checklist and Summary

Use this checklist to determine whether the design exercise is complete. Mark an item complete only after personally collecting the relevant evidence.

  • Identify the OIC instance and training project inspected.
  • Locate Design, Deploy, and Observe in the project workspace.
  • Locate the AI Agents page and inventory its visible resource sections.
  • Distinguish agents, patterns, models and LLM connections, tools, prompt templates, and target-system data.
  • Explain the difference between an ordinary LLM call, a conventional integration, and an agent.
  • Recognize the distinction between OIC Project AI Agents, Fusion AI Agent Studio, OCI Generative AI agent capabilities, and MCP Gateway.
  • Produce the proposed request-path diagram.
  • Complete the read-only Agent Charter with permitted actions, exclusions, data boundaries, and failure behavior.
  • Record runtime scenarios to execute after implementation.
  • Confirm that no unplanned OIC assets or configurations were modified.

The central design principle is straightforward: an agent interprets and coordinates; an integration executes its configured work; and the authorized source system remains responsible for the business data and its actual state.

Once the workspace inventory and charter are ready, the natural next step is to design the read-only integration contract and understand what an OIC integration must satisfy before it can be registered as an agentic AI tool. That implementation belongs to a subsequent module.

10. Five Useful FAQs

FAQ 1 — Does an AI agent replace my existing OIC integrations?

No. Integrations remain responsible for configured connectivity, mappings, validations, orchestration, and system operations. An agent can coordinate eligible registered tools when a request benefits from model-guided interpretation. It does not eliminate the need for deterministic integration design.

FAQ 2 — Is the agent pattern the same as the LLM model?

No. The pattern defines the approach used to coordinate reasoning and actions; the model is selected through the relevant LLM connection. These are related configuration elements with different responsibilities. Confirm available patterns and provider options in your actual OIC environment before implementation.

FAQ 3 — Can an agent retrieve a purchase order when no suitable tool exists?

Not through instructions alone. It needs an available, suitable information path, such as a registered tool backed by an authorized retrieval capability. Without that capability, it must not present a guessed status or amount as an actual procurement record.

FAQ 4 — If I tell the agent to be read-only, is that enough to prevent updates?

No. Instructions influence model behavior but do not independently enforce backend permissions. The implemented tool set, integration design, authorization controls, and target-system permissions must prevent excluded operations. Read-only behavior must be enforced, not merely described.

FAQ 5 — If OIC has an MCP Gateway, why does an agent need registered tools?

They address different design concerns. Registered agentic tools provide capabilities that an OIC Project AI Agent can use. The MCP Gateway provides a governed access path for MCP clients to reach managed MCP servers and their approved capabilities. An MCP Gateway is not automatically required for every agent-tool invocation, and its configuration is outside this lesson's scope.

11. References and Further Reading

Use the documentation for your actual release and the exact task being implemented.

  1. Tour of the Projects Workspace — project navigation, Design/Deploy/Observe areas, AI Agents sections, Knowledge Base, and agent observations.
  2. Design and Build AI Agents — Oracle Integration's agent-development documentation entry point.
  3. Plan Your AI Agent — purpose, tool requirements, and input/output planning.
  4. Configure the Agentic AI Thinking Pattern — pattern concepts and documented configuration.
  5. Register an Integration as an Agentic AI Tool — prerequisites, tool metadata, descriptions, guidelines, and parameter configuration.
  6. Run and Test the AI Agent — configured agent testing and runtime evidence.
  7. September 2026 (26.10) Release Notes — release-specific context, including MCP Gateway.
  8. Overview of AI Agent Studio — Fusion Cloud Applications agent-development context.
  9. Enterprise AI Agents in OCI Generative AI — OCI agent approaches, APIs, and hosted agentic applications.


Comments