A useful AI agent needs more than a good prompt. It needs access to the right information.

An employee might ask about a benefits policy stored in SharePoint. A partner might need guidance from a product website. Another user might simply want to find a document and check when it was last updated.

These requests sound similar, but they need different information—and sometimes different ways of getting it. That’s where connectors come in.

In Oracle Fusion AI Agent Studio, connectors connect your agents and workflows to content sources and external services. This page introduces the building blocks and brings together practical guides you can follow as you explore each connector.

One connection, different possibilities

Before choosing a connector, start with a simple question: What does the agent need to do?

If it needs to answer questions from a collection of documents, an indexed-content approach may be the right fit. Content Intelligence synchronizes and indexes approved content so the workflow can retrieve relevant passages.

If it needs to find a file or check its current properties, a live connector can call the external service when the workflow runs. Depending on the connector, other functions may also support actions such as creating or updating information.

Consider these two requests:

  • “What are the requirements for parental leave?” The agent needs the policy’s contents.
  • “Where can I find the parental-leave policy?” A matching filename and link may be enough.

The difference matters. Finding a document is not the same as reading it.

How the pieces fit together

Three building blocks bring a connected workflow together. Each has its own guide below, so you can explore the connection, the functions, and the workflow step at your own pace.

Component Think of it this way
Connector The connection to your source or service, with its authentication and configuration.
Connector tool The selection of functions you make available to an agent or workflow.
Tool node The workflow step that calls one of those functions with the required inputs.

A simple workflow can start with just a Tool node and an LLM node:

User asks → Tool retrieves information → LLM presents the response

The tool might return document passages, matching files, or structured data. The LLM’s job is to respond using that evidence—not fill in information the tool didn’t provide.

Find the right connector for your use case

The guides below take you from setup to a working example, with screenshots, sample inputs, and checkpoints along the way. This catalog will grow as we add more connectors.

Connector What it brings to your agent Try it when you want to… Explore the guide
WebCrawler Approved website content, synchronized and indexed through Content Intelligence. Answer questions from a product website or learning path. WebCrawler: Answer questions from an approved website
SharePointO365 SharePoint documents, synchronized and indexed through Content Intelligence. Answer employee questions using passages from approved policy documents. SharePointO365: Answer questions from indexed policy documents
Microsoft SharePoint Live Microsoft Graph operations, including file search and metadata retrieval, without CI ingestion in this pattern. Find a document, share its link, or check its last-modified date. Microsoft SharePoint: Find policy files through live Graph operations

Each link opens the corresponding example in our practical workflow guide. Check each example for release-specific requirements and supported functions.

Start with one useful task

You don’t need to enable every function or connect an entire document library to get started. Choose a small, approved source and one task you can easily verify.

For example, retrieve a passage from a known policy or search for a file you know exists. Inspect the tool’s output before adding the LLM response. Then try a request with no matching information and confirm that the workflow handles it clearly.

For indexed connectors, check that synchronization, indexing, and retrieval work. For live connectors, check the connection, permissions, inputs, and response. In both cases, validate access for the people who will use the agent.

Once that first task works end to end, you have a foundation to build on.

Where to go next

Choose a connector from the table and follow its practical example. Each walkthrough focuses on a small, understandable workflow you can test and adapt to your own use case.

For the official setup documentation, start with Create a Connector for AI Agents.

Learn more