Summary
RubyLLM provides three AI inference APIs beyond chat completions — RubyLLM.embed, RubyLLM.paint, and RubyLLM.transcribe — none of which are instrumented by this SDK. The current RubyLLM integration only instruments Chat#complete and Chat#execute_tool.
What is missing
1. Embeddings — RubyLLM::Embedding.embed / RubyLLM.embed
Creates vector embeddings from text. Returns an Embedding object with vectors, model, and input_tokens. Supports provider/model selection and dimensions parameter. This is a core inference API used for RAG and semantic search.
2. Image generation — RubyLLM::Image.paint / RubyLLM.paint
Generates images from text prompts. Returns an Image object with url, data, mime_type, revised_prompt, model_id. Supports model/provider/size selection.
3. Audio transcription — RubyLLM::Transcription.transcribe / RubyLLM.transcribe
Converts audio files to text. Returns a Transcription object with text, model, language, duration, segments, input_tokens, output_tokens. Has built-in token tracking useful for cost attribution.
All three are stable public APIs available since RubyLLM 1.8+ (the minimum version this SDK targets). Each is a single-method call with a simple request/response pattern (no streaming), making instrumentation straightforward.
Braintrust docs status
not_found — The Braintrust RubyLLM integration docs at https://www.braintrust.dev/docs/integrations/ai-sdks/ruby-llm describe support for chat completions across multiple providers and tool/function call tracing but do not mention embeddings, image generation, or transcription.
Upstream sources
Local repo files inspected
lib/braintrust/contrib/ruby_llm/patcher.rb — defines only ChatPatcher, no EmbeddingPatcher, ImagePatcher, or TranscriptionPatcher
lib/braintrust/contrib/ruby_llm/instrumentation/chat.rb — only wraps Chat#complete and Chat#execute_tool
lib/braintrust/contrib/ruby_llm/integration.rb — registers only the ChatPatcher
Summary
RubyLLM provides three AI inference APIs beyond chat completions —
RubyLLM.embed,RubyLLM.paint, andRubyLLM.transcribe— none of which are instrumented by this SDK. The current RubyLLM integration only instrumentsChat#completeandChat#execute_tool.What is missing
1. Embeddings —
RubyLLM::Embedding.embed/RubyLLM.embedCreates vector embeddings from text. Returns an
Embeddingobject withvectors,model, andinput_tokens. Supports provider/model selection and dimensions parameter. This is a core inference API used for RAG and semantic search.2. Image generation —
RubyLLM::Image.paint/RubyLLM.paintGenerates images from text prompts. Returns an
Imageobject withurl,data,mime_type,revised_prompt,model_id. Supports model/provider/size selection.3. Audio transcription —
RubyLLM::Transcription.transcribe/RubyLLM.transcribeConverts audio files to text. Returns a
Transcriptionobject withtext,model,language,duration,segments,input_tokens,output_tokens. Has built-in token tracking useful for cost attribution.All three are stable public APIs available since RubyLLM 1.8+ (the minimum version this SDK targets). Each is a single-method call with a simple request/response pattern (no streaming), making instrumentation straightforward.
Braintrust docs status
not_found— The Braintrust RubyLLM integration docs at https://www.braintrust.dev/docs/integrations/ai-sdks/ruby-llm describe support for chat completions across multiple providers and tool/function call tracing but do not mention embeddings, image generation, or transcription.Upstream sources
Local repo files inspected
lib/braintrust/contrib/ruby_llm/patcher.rb— defines onlyChatPatcher, noEmbeddingPatcher,ImagePatcher, orTranscriptionPatcherlib/braintrust/contrib/ruby_llm/instrumentation/chat.rb— only wrapsChat#completeandChat#execute_toollib/braintrust/contrib/ruby_llm/integration.rb— registers only theChatPatcher