This is an unofficial Java client library that helps to connect your Java applications with OpenAI API. It was developed by AI for Java, but is no longer maintained. This is a fork of the original library, with some new features by Banma Network developers.
- Support Alibaba Qwen-omni models with audio and video as input
- Support enabled / disable Alibaba Qwen3 thinking mode using
extra_body(like what's in python client) - Support run Prompty
- Support Deepseek-R1 model reasoning content
- Added
request_idin ChatCompletionRequest for tracing support - Added
BUILD_FUNCTIONfor kimi model support
- Detailed javadocs
- The rest of API endpoints
- Tell us what you need
Maven:
<dependency>
<groupId>com.ebanma</groupId>
<artifactId>openai4j</artifactId>
<version>0.30.0</version>
</dependency>
Gradle:
implementation 'com.ebanma:openai4j:0.30.0'
Simple way:
String apiKey = System.getenv("OPENAI_API_KEY");
OpenAiClient client = OpenAiClient.builder()
.openAiApiKey(apiKey)
.build();
Customizable way:
String apiKey = System.getenv("OPENAI_API_KEY");
OpenAiClient client = OpenAiClient.builder()
.baseUrl(baseUrl)
.openAiApiKey(apiKey)
.organizationId(orgId)
.callTimeout(ofSeconds(60))
.connectTimeout(ofSeconds(60))
.readTimeout(ofSeconds(60))
.writeTimeout(ofSeconds(60))
.proxy(HTTP, "XXX.XXX.XXX.XXX", 8080)
.logRequests()
.logResponses()
// other customizations coming soon!
.build();
Simple way:
String completion = client.completion("Write a poem about ChatGPT").execute();
Customizable way:
CompletionRequest request = CompletionRequest.builder()
.model(GPT_3_5_TURBO_INSTRUCT)
.prompt("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
CompletionResponse response = client.completion(request).execute();
Simple way:
client.completion("Write a poem about ChatGPT")
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Customizable way:
CompletionRequest request = CompletionRequest.builder()
.model(GPT_3_5_TURBO_INSTRUCT)
.prompt("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
client.completion(request)
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Simple way:
client.completion("Write a poem about ChatGPT")
.onPartialResponse(partialResponse -> ...)
.onComplete(() -> ...)
.onError(error -> ...)
.execute();
Customizable way:
CompletionRequest request = CompletionRequest.builder()
.model(GPT_3_5_TURBO_INSTRUCT)
.prompt("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
client.completion(request)
.onPartialResponse(partialResponse -> ...)
.onComplete(() -> ...)
.onError(error -> ...)
.execute();
Simple way:
String completion = client.chatCompletion("Write a poem about ChatGPT").execute();
Customizable way:
ChatCompletionRequest request = ChatCompletionRequest.builder()
.model(GPT_3_5_TURBO)
.addSystemMessage("You are a helpful assistant")
.addUserMessage("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
ChatCompletionResponse response = client.chatCompletions(request).execute();
Simple way:
client.chatCompletion("Write a poem about ChatGPT")
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Customizable way:
ChatCompletionRequest request = ChatCompletionRequest.builder()
.model(GPT_3_5_TURBO)
.addSystemMessage("You are a helpful assistant")
.addUserMessage("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
client.chatCompletion(request)
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Simple way:
client.chatCompletion("Write a poem about ChatGPT")
.onPartialResponse(partialResponse -> ...)
.onComplete(() -> ...)
.onError(error -> ...)
.execute();
Customizable way:
ChatCompletionRequest request = ChatCompletionRequest.builder()
.model(GPT_3_5_TURBO)
.addSystemMessage("You are a helpful assistant")
.addUserMessage("Write a poem about ChatGPT")
.temperature(0.9)
...
.build();
client.chatCompletion(request)
.onPartialResponse(partialResponse -> ...)
.onComplete(() -> ...)
.onError(error -> ...)
.execute();
Simple way:
List<Float> embedding = client.embedding("Write a poem about ChatGPT").execute();
Customizable way:
EmbeddingRequest request = EmbeddingRequest.builder()
.model(TEXT_EMBEDDING_ADA_002)
.input("Write a poem about ChatGPT", "Write a haiku about ChatGPT")
...
.build();
EmbeddingResponse embedding = client.embedding(request).execute();
Simple way:
client.embedding("Write a poem about ChatGPT")
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Customizable way:
EmbeddingRequest request = EmbeddingRequest.builder()
.model(TEXT_EMBEDDING_ADA_002)
.input("Write a poem about ChatGPT", "Write a haiku about ChatGPT")
...
.build();
client.embedding(request)
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Simple way:
ModerationResult moderationResult = client.moderation("Write a poem about ChatGPT").execute();
Customizable way:
ModerationRequest request = ModerationRequest.builder()
.input(INPUT)
.model(TEXT_MODERATION_STABLE)
.build();
ModerationResponse response = client.moderation(request).execute();
Simple way:
client.moderation("Write a poem about ChatGPT")
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Customizable way:
ModerationRequest request = ModerationRequest.builder()
.input(INPUT)
.model(TEXT_MODERATION_STABLE)
.build();
client.moderation(request)
.onResponse(response -> ...)
.onError(error -> ...)
.execute();
Simple way:
OpenAiClient client = OpenAiClient
.builder()
.openAiApiKey(System.getenv("OPENAI_API_KEY"))
.build();
ImageRequest request = ImageRequest
.builder()
.prompt("Beautiful house on country side")
.build();
ImageResponse response = client.imagesGenerations(request).execute();
// remote image generated with model DALL·E 3 and resolution 1024x1024
String remoteImage = response.data().get(0).url();
Customizable way:
OpenAiClient client = OpenAiClient
.builder()
.openAiApiKey(System.getenv("OPENAI_API_KEY"))
.logRequests()
.logResponses()
.withPersisting()
.build();
ImageRequest request = ImageRequest
.builder()
.model(DALL_E_3)
.size(DALL_E_SIZE_1792_x_1024)
.quality(DALL_E_QUALITY_HD)
.responseFormat(DALL_E_RESPONSE_FORMAT_B64_JSON) // if you like to get the image within the response body
.prompt("Cute red parrot flying in the sky")
.build();
ImageResponse response = client.imagesGenerations(request).execute();
// your generated image is here locally:
String localImage = response.data().get(0).url();
Audio Support
OpenAiClient client = OpenAiClient.builder()
.baseUrl("https://dashscope.aliyuncs.com/compatible-mode/v1")
.openAiApiKey("<your api key>")
.logRequests()
.logResponses()
.build();
List<Message> messages = new ArrayList<>();
CustomMessage sys = CustomMessage.builder().role(Role.SYSTEM).addContent(Content.builder().type(ContentType.TEXT).text("You are a helpful assistant.").build()).build();
messages.add(sys);
InputAudio audio = InputAudio.builder()
.data("https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250211/tixcef/cherry.wav")
.format("wav")
.build();
CustomMessage user = CustomMessage.builder().role(Role.USER)
.addContent(Content.builder().type(ContentType.INPUT_AUDIO).inputAudio(audio).build())
.addContent(Content.builder().type(ContentType.TEXT).text("这段音频在说什么？").build())
.build();
messages.add(user);
ChatCompletionRequest request = ChatCompletionRequest.builder()
.model("qwen-omni-turbo")
.stream(true)
.modalities(asList("text", "audio"))
.audio(audio())
.messages(messages)
.requestId(UUID.randomUUID().toString())
.build();
client.chatCompletion(request)
See more examples with image, audio and video in src/test/java/dev/ai4j/openai4j/chat/QwenOmniTest.java
Disable Thinking
OpenAiClient client = OpenAiClient.builder()
.baseUrl("https://dashscope.aliyuncs.com/compatible-mode/v1")
.openAiApiKey("<your api key>")
.logRequests()
.logResponses()
.build();
Map<String, Object> extraBody = new HashMap<>();
extraBody.put("enable_thinking", Boolean.FALSE);
ChatCompletionRequest request = ChatCompletionRequest
.builder()
.model("qwen3-8b")
.requestId(UUID.randomUUID().toString().replace("-", ""))
.stream(true)
.extraBody(extraBody)
.messages(UserMessage.from("你是谁？"))
.build();
StringBuilder content = new StringBuilder();
StringBuilder reasoning_content = new StringBuilder();
client.chatCompletion(request)
See more examples in src/test/java/dev/ai4j/openai4j/chat/QwenThinkingTest.java
- How to get best results form AI: https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/
- Text version of above course: https://platform.openai.com/docs/guides/gpt-best-practices
- How to build software powered by OpenAI/ChatGPT: https://www.deeplearning.ai/short-courses/building-systems-with-chatgpt/
- Cookbook with examples of how to use OpenAI API: https://github.com/openai/openai-cookbook
- ChatGPT DALL-E 3: Complete Guide (Generate Images with Text) https://www.akkio.com/post/chatgpt-dall-e-3