[bot] Spring AI OpenAI auto-instrumentation only covers chat — Image/Audio (speech & transcription) models are not instrumented

SpringAIOpenAIInstrumentationModule.java の39–58行目と BraintrustSpringAI.java から始め、次に OpenAiImageModel、OpenAiAudioTranscriptionModel、OpenAiAudioSpeechModel の実行APIを調査します。画像、文字起こし、音声呼び出しのカバレッジを BraintrustSpringAITest.java に追加します。これらのモデルが spans を生成し、既存の chat と stream のテストが引き続き成功すれば完了です。

<!-- provider-gap-audit: springai-openai-image-audio-not-instrumented -->

Summary

SpringAIOpenAIInstrumentationModule only installs a ByteBuddy hook on org.springframework.ai.openai.OpenAiChatModel$Builder.build(). Spring AI's spring-ai-openai artifact also ships stable, documented execution-model classes for image generation (OpenAiImageModel) and audio (OpenAiAudioTranscriptionModel, OpenAiAudioSpeechModel) — none of these are targeted by any TypeInstrumentation, so calls made through them produce no spans at all (not even generic/untagged ones, since the interception point itself never fires).

This is distinct from the already-filed embedding gap ("Spring AI instrumentation limited to OpenAI and Anthropic chat models; no embedding model support"), which covers OpenAiEmbeddingModel. It's also distinct from the direct-OpenAI-SDK gaps ("OpenAI Images API spans missing input/output capture", "OpenAI audio transcription and speech spans missing braintrust.output_json"), which are about com.openai:openai-java's own client.images()/client.audio(), not the Spring AI wrapper.

What is missing

braintrust-sdk/instrumentation/springai_1_0_0/src/main/java/dev/braintrust/instrumentation/springai/v1_0_0/auto/SpringAIOpenAIInstrumentationModule.java:

@Override
public List<TypeInstrumentation> typeInstrumentations() {
    return List.of(new OpenAiChatModelBuilderInstrumentation());
}

Only OpenAiChatModelBuilderInstrumentation (matching org.springframework.ai.openai.OpenAiChatModel$Builder) is registered. There is no equivalent TypeInstrumentation for:

• org.springframework.ai.openai.OpenAiImageModel — image generation execution API (ImageModel.call())
• org.springframework.ai.openai.OpenAiAudioTranscriptionModel — speech-to-text execution API
• org.springframework.ai.openai.OpenAiAudioSpeechModel — text-to-speech execution API

All three are constructed independently of OpenAiChatModel and make their own HTTP calls through Spring AI's OpenAiAudioApi/OpenAiImageApi, which are not wrapped by BraintrustRestInterceptor/BraintrustWebClientFilter the way OpenAiApi (used by chat) is. Manual wrapping via BraintrustSpringAI also has no entry point for these model types — grep for ImageModel, AudioTranscriptionModel, AudioSpeechModel in braintrust-sdk/instrumentation/springai_1_0_0/ returns zero matches.

Braintrust docs status

not_found — the Braintrust docs for Spring AI / Java integrations do not mention image or audio model coverage; only chat completion examples are shown.

Upstream sources

• Spring AI OpenAI Image generation: https://docs.spring.io/spring-ai/reference/api/image/openai-image.html
• Spring AI OpenAI Audio Transcription: https://docs.spring.io/spring-ai/reference/api/audio/transcriptions/openai-transcriptions.html
• Spring AI OpenAI Text-to-Speech: https://docs.spring.io/spring-ai/reference/api/audio/speech/openai-speech.html
• OpenAiAudioTranscriptionModel API reference: https://docs.spring.io/spring-ai/docs/current/api/org/springframework/ai/openai/OpenAiAudioTranscriptionModel.html

Local files inspected

• braintrust-sdk/instrumentation/springai_1_0_0/src/main/java/dev/braintrust/instrumentation/springai/v1_0_0/auto/SpringAIOpenAIInstrumentationModule.java — lines 39–58 (only chat model builder instrumentation registered)
• braintrust-sdk/instrumentation/springai_1_0_0/src/main/java/dev/braintrust/instrumentation/springai/v1_0_0/BraintrustSpringAI.java — no image/audio model wrap entry point
• braintrust-sdk/instrumentation/springai_1_0_0/src/test/java/dev/braintrust/instrumentation/springai/v1_0_0/BraintrustSpringAITest.java — only chat model tests (OpenAI + Anthropic, call/stream)
• Full-module grep for ImageModel, AudioTranscriptionModel, AudioSpeechModel — zero matches