OpenAI Official Model

agentscope-extensions-model-openai-official integrates OpenAI models through the official OpenAI Java SDK. It currently supports only the Responses API; support for the Chat Completions API may be added in the future.

Add the dependency

<dependency>
    <groupId>io.agentscope</groupId>
    <artifactId>agentscope-extensions-model-openai-official</artifactId>
    <version>${agentscope.version}</version>
</dependency>

ModelRegistry

Set OPENAI_API_KEY, then use the openai-official:<model> id:

ReActAgent agent = ReActAgent.builder()
    .name("assistant")
    .model("openai-official:gpt-4o")
    .build();

Explicit builder

Use the builder when you need a custom base URL, or a multi-agent formatter:

import io.agentscope.extensions.model.openaiofficial.OpenAIResponsesChatModel;

OpenAIResponsesChatModel model = OpenAIResponsesChatModel.builder()
    .apiKey(System.getenv("OPENAI_API_KEY"))
    .modelName("gpt-4o")
    .stream(true)
    .build();

Spring Boot

There is no dedicated Spring Boot starter for this module yet.

Reasoning

Control reasoning effort via GenerateOptions.reasoningEffort (low / medium / high / minimal). To make reasoning visible in the response, explicitly enable reasoning.summary:

GenerateOptions options = GenerateOptions.builder()
    .reasoningEffort("high")
    .additionalBodyParam("reasoning.summary", "auto")
    .build();

Encrypted reasoning content from previous turns is automatically replayed in multi-turn conversations via Msg.metadata. No manual management is needed.

Compatibility notes

This module integrates via the OpenAI Java SDK and currently supports only the Responses API; Chat Completions API support may be added in the future. The following options are not supported and will fail-fast when set: frequencyPenalty, presencePenalty, topK, seed, cacheControl, thinkingBudget, endpointPath, per-request additionalHeaders, and per-request additionalQueryParams. Responses-specific parameters are available through GenerateOptions.additionalBodyParams with a whitelist: reasoning.summary, reasoning.context, reasoning.mode, service_tier, prompt_cache_key, prompt_cache_options, max_tool_calls, safety_identifier, store, and previous_response_id. store controls server-side response storage and defaults to false. To use OpenAI’s server-side conversation chaining, store the first response, read its ID from response metadata (openai.response.id), and pass that ID as previous_response_id in a later request. When previous_response_id is used, send only the new turn’s input rather than repeating the full conversation history. Native structured output is always enabled (supportsNativeStructuredOutput() returns true). The SDK retry is disabled (maxRetries=0); retry is managed by AgentScope.

Integration

Channels

Distributed storage

OpenAI Official

OpenAI Official Model

Add the dependency

ModelRegistry

Explicit builder

Spring Boot

Reasoning

Compatibility notes

​OpenAI Official Model

​Add the dependency

​ModelRegistry

​Explicit builder

​Spring Boot

​Reasoning

​Compatibility notes

OpenAI Official Model

Add the dependency

ModelRegistry

Explicit builder

Spring Boot

Reasoning

Compatibility notes