[bot] Missing instrumentation for Azure AI Inference Java SDK

[braintrust-bot]

Summary

The Braintrust Java SDK does not instrument the Azure AI Inference Java SDK (com.azure:azure-ai-inference). This is Microsoft's official Java client for the Azure AI Foundry model catalog and GitHub Models, providing execution-oriented APIs for chat completions (sync and streaming), text embeddings, and function/tool calling. It is a distinct library from com.azure:azure-ai-openai (tracked in #68) — it targets a different Azure service endpoint and uses a different client hierarchy.

There is no instrumentation module, wrapper, test, example, or dependency for azure-ai-inference anywhere in this repository.

What is missing

The com.azure:azure-ai-inference SDK provides these execution surfaces, none of which are instrumented:

API	Class	Description
complete()	ChatCompletionsClient	Synchronous chat completions
completeAsync()	ChatCompletionsAsyncClient	Async chat completions
completeStream()	ChatCompletionsClient	Streaming chat completions (delta chunks)
embed()	EmbeddingsClient	Text embedding generation
Function/tool calling	ChatCompletionsOptions	Tool invocation during model execution
Multimodal	ChatMessageImageContentItem	Image inputs for vision-capable models

Why existing instrumentation does not cover this

The existing openai_2_15_0 module wraps com.openai:openai-java's internal HTTP client. The azure-ai-openai issue (#68) covers com.azure:azure-ai-openai, which targets the Azure OpenAI Service endpoint (/openai/deployments/).

com.azure:azure-ai-inference is a separate library targeting the Azure AI Inference Service endpoint (/models/chat/completions), used for:

• GitHub Models (free-tier inference from github.com)
• Azure AI Foundry model catalog (Llama, Mistral, Cohere, and other non-OpenAI models via Azure)

It uses the com.azure.core HTTP pipeline (Netty-based), which is completely separate from all existing instrumentation. Neither the OpenAI module nor the Azure OpenAI issue covers these calls.

Braintrust docs status

• Azure AI Foundry integration is documented at https://www.braintrust.dev/docs/integrations/ai-providers/azure — but only covers API key configuration; no Java SDK instrumentation is mentioned
• No reference to azure-ai-inference or GitHub Models Java SDK exists in Braintrust docs: not_found

Upstream sources

• Azure AI Inference Java SDK README: https://learn.microsoft.com/en-us/java/api/overview/azure/ai-inference-readme
• GitHub repository: https://github.com/Azure/azure-sdk-for-java/tree/main/sdk/ai/azure-ai-inference
• Maven Central: https://central.sonatype.com/artifact/com.azure/azure-ai-inference — v1.0.0-beta.5 (May 15, 2025), part of the Azure SDK for Java suite
• API reference: https://learn.microsoft.com/en-us/java/api/com.azure.ai.inference

Local files inspected

• braintrust-sdk/instrumentation/ — contains openai_2_15_0, anthropic_2_2_0, genai_1_18_0, langchain_1_8_0, springai_1_0_0, aws_bedrock_2_30_0; no azure-ai-inference directory
• settings.gradle — lists all instrumentation modules; no Azure AI Inference module present
• Full-repo grep for azure.ai.inference, azure-ai-inference, ChatCompletionsClient, EmbeddingsClient (Azure) — zero matches