OpenAI (Java)

Prerequisites

Complete the Java SDK setup first. You need TraceAI.init() called before using this wrapper.

Installation

<dependency>
    <groupId>com.github.future-agi.traceAI</groupId>
    <artifactId>traceai-java-openai</artifactId>
    <version>main-SNAPSHOT</version>
</dependency>

implementation 'com.github.future-agi.traceAI:traceai-java-openai:main-SNAPSHOT'

You also need the OpenAI Java SDK:

<dependency>
    <groupId>com.openai</groupId>
    <artifactId>openai-java</artifactId>
    <version>0.8.0</version>
</dependency>

implementation 'com.openai:openai-java:0.8.0'

Wrap the client

import ai.traceai.TraceAI;
import ai.traceai.openai.TracedOpenAIClient;
import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

// Initialize TraceAI (once, at startup)
TraceAI.initFromEnvironment();

// Create the OpenAI client
OpenAIClient client = OpenAIOkHttpClient.builder()
    .apiKey(System.getenv("OPENAI_API_KEY"))
    .build();

// Wrap it
TracedOpenAIClient traced = new TracedOpenAIClient(client);

Or with an explicit tracer:

import ai.traceai.FITracer;

FITracer tracer = TraceAI.getTracer();
TracedOpenAIClient traced = new TracedOpenAIClient(client, tracer);

Chat completions

import com.openai.models.*;

ChatCompletion response = traced.createChatCompletion(
    ChatCompletionCreateParams.builder()
        .model("gpt-4o-mini")
        .addMessage(ChatCompletionMessageParam.ofChatCompletionSystemMessageParam(
            ChatCompletionSystemMessageParam.builder()
                .role(ChatCompletionSystemMessageParam.Role.SYSTEM)
                .content(ChatCompletionSystemMessageParam.Content.ofTextContent(
                    "You are a helpful assistant."))
                .build()))
        .addMessage(ChatCompletionMessageParam.ofChatCompletionUserMessageParam(
            ChatCompletionUserMessageParam.builder()
                .role(ChatCompletionUserMessageParam.Role.USER)
                .content(ChatCompletionUserMessageParam.Content.ofTextContent(
                    "What is the capital of France?"))
                .build()))
        .temperature(0.7)
        .build()
);

System.out.println(response.choices().get(0).message().content().orElse(""));

Span created: “OpenAI Chat Completion” with kind LLM

Embeddings

import com.openai.models.*;

CreateEmbeddingResponse response = traced.createEmbedding(
    EmbeddingCreateParams.builder()
        .model("text-embedding-3-small")
        .input(EmbeddingCreateParams.Input.ofString("Hello world"))
        .build()
);

System.out.println("Dimensions: " + response.data().get(0).embedding().size());

Span created: “OpenAI Embedding” with kind EMBEDDING

Streaming

The streaming wrapper collects all chunks, records the full response in the span, then returns them as an Iterable:

import com.openai.models.*;

Iterable<ChatCompletionChunk> chunks = traced.streamChatCompletion(
    ChatCompletionCreateParams.builder()
        .model("gpt-4o-mini")
        .addMessage(ChatCompletionMessageParam.ofChatCompletionUserMessageParam(
            ChatCompletionUserMessageParam.builder()
                .role(ChatCompletionUserMessageParam.Role.USER)
                .content(ChatCompletionUserMessageParam.Content.ofTextContent(
                    "Write a haiku about Java."))
                .build()))
        .build()
);

for (ChatCompletionChunk chunk : chunks) {
    chunk.choices().get(0).delta().content().ifPresent(System.out::print);
}

Span created: “OpenAI Chat Completion (Stream)” with kind LLM. The span captures the accumulated full response, not individual chunks.

What gets captured

Chat completion spans

Embedding spans

Accessing the original client

If you need the unwrapped client for operations that aren’t traced:

OpenAIClient original = traced.unwrap();