Area(s)

area:gen-ai

What's missing?

Common nomenclature (or prefixing) for Google's GenAI systems.

This divergent naming can also make it difficult to know which value to populate in gen_ai.system at higher-level abstraction layers that build on top of the Google GenAI SDK (which provides the same library interface, regardless of whether Vertex AI or the Gemini backend is used).

Describe the solution you'd like

Common proposal

When Google is a backend, use the Google-specific gcp.client.service convention to identify the specific backend where known:

• gcp.client.service:
  • value == generativelanguage => generativelanguage.googleapis.com (Gemini)
  • value == aiplatform => aiplatform.googleapis.com (Vertex AI)

Ideal solution

In conjunction with the common proposal above, reduce the two values of gemini and vertex_ai for gen_ai.system to a single value: google

Alternative solution

In conjunction with the common proposal above:

Rename the existing gen_ai.system values for Google:

• gemini => google.gemini
• vertex_ai => google.gcp.vertex

And define an additional value:

• google (for when the specific subsystem -- vertex vs gemini -- is not known)

This way there is at least a common prefix (google) for which shared properties can be attached.