Google GenAI: embed_content not instrumented

[AbhiPrasad]

Gap

The Google GenAI SDK's embed_content method (client.models.embed_content() / client.aio.models.embed_content()) is not instrumented by the Braintrust wrapper. Embeddings are a core AI API surface used in RAG pipelines, semantic search, and clustering — and Braintrust already instruments embeddings for other providers (OpenAI via EmbeddingV1Wrapper, LiteLLM via EmbeddingWrapper).

What is missing

• client.models.embed_content() (sync)
• client.aio.models.embed_content() (async)

These methods accept model, contents, and config (with task_type, output_dimensionality) and return embedding vectors with metadata.

Braintrust docs status

not_found — the Gemini integration page at braintrust.dev/docs/integrations/ai-providers/gemini documents generateContent and streaming but does not mention embeddings.

Upstream sources

• Google GenAI SDK embeddings guide: https://ai.google.dev/gemini-api/docs/embeddings
• google.genai.models.Models.embed_content method in the google-genai Python package

Local files inspected

• py/src/braintrust/wrappers/google_genai/__init__.py — setup_genai() patches Models._generate_content and Models.generate_content_stream (and async variants) but not embed_content. Confirmed via grep: zero references to embed_content or embedding in the file.
• py/src/braintrust/wrappers/test_google_genai.py — no embedding test cases
• py/noxfile.py — test_google_genai session exists but only tests generation