feat(genai,vertexai): expose provider response ids

[Ayush Srivastava (luigivendetta)]

Description

Expose Google's provider response identifier as AIMessage.response_metadata["id"] for both langchain-google-genai and langchain-google-vertexai.

This lets users correlate LangChain responses with Gemini and Vertex dashboards/logs without storing prompt or response text. The metadata key matches langchain-openai, which already exposes the native provider response ID through response_metadata["id"].

This intentionally stores the provider ID in response_metadata, not message.id, so it remains additive and does not change LangChain message identity behavior.

Relevant issues

N/A

Type

🆕 New Feature
✅ Test

Changes

• Add GenAI non-stream response ID metadata from GenerateContentResponse.response_id.
• Add GenAI streaming response ID metadata on the final chunk only.
• Add Vertex response ID metadata for non-streaming responses.
• Add Vertex fallback extraction for raw response shapes that expose response_id / responseId.
• Add Vertex streaming response ID metadata on the final chunk only.
• Preserve existing message.id, additional_kwargs, content parsing, thinking blocks, tool-call IDs, and usage metadata behavior.

Why this approach

• Uses response_metadata["id"] for provider parity with langchain-openai.
• Keeps the change backward compatible by only adding metadata when the provider returns an ID.
• Attaches streaming IDs only to the final chunk so concatenated streamed messages contain the ID once.
• Keeps GenAI and Vertex behavior consistent across Google chat integrations.

Review notes

Areas worth extra review:

• Whether response_metadata["id"] is the preferred cross-provider key.
• Whether keeping the ID only on the final streaming chunk is the right merge behavior.
• Whether the Vertex fallback extraction covers the expected response shapes without being too broad.

Testing

Passed locally:

• cd libs/genai && UV_CACHE_DIR=/tmp/uv-cache uv lock --check
• cd libs/genai && UV_CACHE_DIR=/tmp/uv-cache make lint_package
• cd libs/genai && UV_CACHE_DIR=/tmp/uv-cache make lint_tests
• cd libs/genai && UV_CACHE_DIR=/tmp/uv-cache make test TEST_FILE=tests/unit_tests/test_chat_models.py
  • 217 passed
• cd libs/genai && UV_CACHE_DIR=/tmp/uv-cache uv run --group test --group test_integration pytest -m compile tests/integration_tests
• cd libs/vertexai && UV_CACHE_DIR=/tmp/uv-cache uv lock --check
• cd libs/vertexai && UV_CACHE_DIR=/tmp/uv-cache make lint_package
• cd libs/vertexai && UV_CACHE_DIR=/tmp/uv-cache make lint_tests
• cd libs/vertexai && UV_CACHE_DIR=/tmp/uv-cache make test TEST_FILE=tests/unit_tests/test_chat_models.py
  • 110 passed, 1 skipped
• cd libs/vertexai && UV_CACHE_DIR=/tmp/uv-cache uv run --group test --group test_integration pytest -m compile tests/integration_tests

Note

This touches both GenAI and Vertex intentionally so provider response ID metadata lands consistently across Google chat integrations. I can split this into separate package PRs if that is easier to review.

This contribution was prepared with assistance from AI agents. The author reviewed and validated the changes locally.