OCI Responses API

OCI Responses API is the OCI-native API for agentic workflows.

Installation

Python

Install the OCI GenAI Auth Python package:

pip install oci-genai-auth

Java

See OCI GenAI Auth Java SDK on GitHub .

The OCI Responses API lets you compose modular primitives to support agentic workloads. You can configure these capabilities at the request level:

Orchestration to perform single-step inference or multi-step agentic workflows
Reasoning to enable or disable reasoning
Tools to use platform-managed tools, client-side function tools, or both
Conversation state to manage state on the service side or in your client application

This approach provides:

Flexibility to build simple assistants or more complex multi-step agents by using the same API
Precision to control cost, latency, and behavior
Simplicity because you do not need to create and manage tools as separate cloud resources

Use the OCI Responses API at the following base URL:

https://inference.generativeai.<available-region>.oci.oraclecloud.com/openai/v1

Replace <available-region> with a region identifier such as generativeai.us-chicago-1.

You can use the OCI Responses API with the OpenAI SDK. You can also use it with compatible client-side agent frameworks.

The OpenAI SDK supports these languages:

More language support is available through community libraries.

Compatible agent frameworks include:


Dimension	OCI Responses API	OCI Chat API using Chat Completions API
Primary use	Unified API for model interaction and agentic capabilities	API for model interaction
Best fit	Interactive chat, agentic workloads, and long-running tasks	Interactive chatbots and text completion
Orchestration	Built-in multi-step reasoning and multiple tool calls	Single-step inference or generation; multi-step flows require external orchestration
Context management	Stateful by default, with optional stateless usage	Stateless only; the client manages conversation history
Tool support	Built-in tools such as File Search, Code Interpreter, and remote MCP	Limited to client-side tools through function calling
Multimodal support	Native support for text, images, files, and structured content blocks	Primarily text, with limited multimodal support depending on the model version
Streaming	Event-based streaming with fine-grained events	Token-based streaming
Structured output	Native structured outputs and JSON schema enforcement	JSON mode is supported, but is less composable
File and vector integration	Direct integration with Files API and Vector Stores API	Requires separate orchestration
Extensibility	Designed for hosted agents, MCP tools, memory, and containers	Designed primarily for chat applications


Capability	OCI Responses API	OpenAI Responses API
Model choice	Supports OCI-hosted models and non-OpenAI models	OpenAI models only
Model serving infrastructure	OCI shared infrastructure or dedicated AI clusters	OpenAI shared infrastructure
Authentication	OCI IAM or API keys	API keys
Data retention	Your data remains in OCI	Your data is stored in OpenAI
Private networking	Supports OCI VCN integration and private endpoints	Not available
Endpoint model	Regional endpoints	Global endpoint

Use the OpenAI SDK when:

You do not need a client-side framework to orchestrate function tools
Your tools run on the server side, including remote MCP tools
You use the Conversations API to manage session memory
Your application is a simpler interface, such as a mobile app or static site, without a complex backend

Use an agent framework with the OpenAI SDK when:

You need to orchestrate client-side function tools
You need client-side execution capabilities such as local shell or file system access
You need to coordinate multiple agents on the client side
You manage session memory by using your own solution or a third-party system