Enterprise AI Agents in OCI Generative AI

For development, OCI Generative AI provides managed APIs that you can use to build and run agentic workloads directly from your application.

The main API for building agents is the OCI Responses API, which is an OpenAI-compatible, Open Responses-complian for interacting with models and agentic capabilities across providers.

You can build agents directly through the OCI Responses API by using modular capabilities such as:

• Agent Tools, including Web Search, File Search, Code Interpreter, Image Generation, Function Calling, and MCP Calling
• Agent Memory, including the Conversations API, Long-Term Memory, and Memory Context Optimization
• Lower-level agent building blocks, including the Vector Stores API, Files API, and Containers API

You can use these managed APIs to build and run agentic workloads without deploying any infrastructure. In the simplest case, your application invokes the OCI Responses API directly and composes the tools, memory, and other resources that your workload requires.

For more information, see the topics about the OCI Responses API, agent tools, agent memory, and lower-level agent building blocks.

For deployment, OCI Generative AI provides hosted applications that let you package and run agentic workloads as managed services.

Hosted applications support:

• Agentic Applications Hosting for deployment and auto-scaling of agents built with frameworks such as OpenAI Agents SDK, LangChain, LangGraph, and AutoGen
• MCP Servers Hosting for deployment and auto-scaling of MCP servers

To deploy an agent, you build a container image, upload it to Oracle Cloud Infrastructure Registry (OCIR), and deploy it as a hosted application by using the OCI Console, API, or CLI.

When you deploy an application, you configure settings such as scaling, storage, networking, and authentication. After the deployment is provisioned, OCI Generative AI provides an endpoint that clients can use to invoke the deployed agent.

For more information, see the topics about hosted applications and deploying agentic workloads.

The following diagram shows both the building blocks used to build agents with the OCI Responses API and the deployment components used to run agentic workloads on OCI Generative AI.

On the left side, the diagram shows an agentic application, including model SDKs, agent frameworks, prompts, model configurations, and locally defined tools and hooks. In the center, it shows the OCI Responses API (shown as Agentic API) and related resources, including agent tools, memory, Files API, Vector Stores API, Containers API, and Container Files API. On the right side, it shows the managed runtime and deployment-side components, including scalable platform infrastructure, model servers, tool servers, secure sandboxes, and third-party services.

• IAM permissions:
  • Give user groups access to all Generative AI resources.
  • Give OCI resources access to other OCI resources:
    • For OCI Generative AI Vector Store Connectors
    • For OCI Generative AI Applications
    • For OCI Vulnerability Scanning Service
• Set up authentication for agentic support..
• Install the OCI GenAI Auth package.
• Create the OCI Generative AI resources needed for your Enterprise AI task:
  1. Create a project.
  2. Create an API key.
  3. Add key permissions.
  4. Create an application.
  5. Create a deployment.
  6. Add an artifact.
  7. Create a vector store.
• Complete the next step based on the type of vector store:
  • For structured data, use the NL2SQL API.
  • For unstructured data, sync data to the vector store.