Add support for GenAI instrumentation

[xuan-cao-swi]

We probably should start to support some of the GenAI instrumentation unless they have native instrumentation.

gem 'openai'                           # Official OpenAI API client
gem 'ruby-openai'                      # OpenAI API client
gem 'langchainrb'                      # LangChain for Ruby  (it's using unmaintained third party llm lib)
gem 'ruby_llm'                         # LangChain-like library (actively maintained)
gem 'google-cloud-ai_platform'         # Google Vertex AI
gem 'gemini-ai'                        # Google Gemini AI (maybe should focus on the official google-cloud-ai_platform)

Spec: https://opentelemetry.io/docs/specs/semconv/gen-ai/

[github-actions]

👋 This issue has been marked as stale because it has been open with no activity. You can: comment on the issue or remove the stale label to hold stale off for a while, add the keep label to hold stale off permanently, or do nothing. If you do nothing this issue will be closed eventually by the stale bot.

[aviralgarg05]

I would like to solve this issue, please assign me

[xuan-cao-swi]

Thanks @aviralgarg05 , the openai gem has the PR. Which llm lib you want to tackle first?

[aviralgarg05]

Thanks for raising this! Supporting GenAI instrumentation is definitely a key addition given the increasing adoption.
I agree that we should prioritize the official openai gem first. Since strict adherence to the GenAI Semantic Conventions is important here (capturing attributes like gen_ai.request.model, gen_ai.usage.input_tokens, etc.), a dedicated instrumentation wrapping the OpenAI::Client methods (like chat and completions) would likely be cleaner than trying to rely on the generic net/http instrumentation alone.
I'm interested in looking into this.

Since I don't currently have a paid API key for live verification, I plan to rely heavily on VCR/WebMock to simulate the API responses and ensure the spans and attributes match the spec. This approach should ensure the instrumentation is robust even without live integration tests running in every CI build.

[arielvalentin]

Is there anything we can do to engage SDK teams directly first to see if they are willing to support first part support for otel instrumentation before adding them to the contrib repo?

[xuan-cao-swi]

We could raise an issue or submit a pull request in their official repository to request support for the OpenTelemetry SDK. However, it's unclear how committed they might be to this effort.

[github-actions]

👋 This issue has been marked as stale because it has been open with no activity. You can: comment on the issue or remove the stale label to hold stale off for a while, add the keep label to hold stale off permanently, or do nothing. If you do nothing this issue will be closed eventually by the stale bot.