Python: fix(google-ai): skip api_key check when use_vertexai is True

[giulio-leone]

Summary

Fixes #13483

Motivation and Context

When use_vertexai=True, users authenticate via Application Default Credentials (ADC) and don't need an API key. However, the initialization check for api_key runs unconditionally, blocking Vertex AI users with:

ServiceInitializationError: The API key is required when use_vertexai is False.

Description

Guard the api_key validation with not google_ai_settings.use_vertexai in all three Google AI service classes:

• GoogleAIChatCompletion
• GoogleAITextCompletion
• GoogleAITextEmbedding

The error message already reads "when use_vertexai is False", so the condition now matches the intent.

Before / After

# Before — fails even with use_vertexai=True
GoogleAIChatCompletion(
    use_vertexai=True,
    cloud_project_id="my-project",
    cloud_region="us-central1"
)
# → ServiceInitializationError: The API key is required when use_vertexai is False.

# After — works correctly
GoogleAIChatCompletion(
    use_vertexai=True,
    cloud_project_id="my-project",
    cloud_region="us-central1"
)
# → Initializes successfully, uses ADC for authentication

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines
• The code follows the SK coding standards

[giulio-leone]

@microsoft-github-policy-service agree

[giulio-leone]

Friendly ping — CI is green and this is ready for review. Happy to address any feedback. Thanks!

[Copilot]

Pull request overview

This PR fixes Google AI connector initialization for Vertex AI usage by skipping the api_key requirement when use_vertexai=True, aligning the validation logic with the intended ADC authentication path.

Changes:

• Guard api_key validation with not use_vertexai in GoogleAIChatCompletion.
• Guard api_key validation with not use_vertexai in GoogleAITextCompletion.
• Guard api_key validation with not use_vertexai in GoogleAITextEmbedding.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
python/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_chat_completion.py	Skip API key validation when using Vertex AI during initialization.
python/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_completion.py	Skip API key validation when using Vertex AI during initialization.
python/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_embedding.py	Skip API key validation when using Vertex AI during initialization.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[giulio-leone]

Hi! Gentle ping — this PR is rebased, CI passes, and ready for review. Happy to address any feedback. Thanks!

[giulio-leone]

Friendly ping — CI is green, tests pass, ready for review whenever convenient. Happy to address any feedback. Thanks! 🙏

[giulio-leone]

Hi — all 4 threads resolved. cloud_region validation added to all 3 service classes. Fixes #13483. Ready for review!

[giulio-leone]

Rebased this branch onto current main (bb421f67a, PR #13686) and reran local validation twice with no code changes between passes.

Local validation (Pass 1 and Pass 2)

• UV_PYTHON=3.10 uv run --frozen ruff check semantic_kernel/connectors/ai/google/google_ai/services/google_ai_chat_completion.py semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_completion.py semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_embedding.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py
  • both passes: All checks passed!
• UV_PYTHON=3.10 uv run --frozen pytest tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py
  • both passes: 53 passed

Direct runtime proof (real source files, not mocks)

Using the real origin/main source tree vs this rebased branch with:

• GOOGLE_AI_GEMINI_MODEL_ID=test-gemini-model-id
• GOOGLE_AI_EMBEDDING_MODEL_ID=test-embedding-model-id
• GOOGLE_AI_CLOUD_PROJECT_ID=test-project-id
• GOOGLE_AI_CLOUD_REGION=test-region
• no GOOGLE_AI_API_KEY

Results were identical across both proof passes:

Case	Chat	Text	Embedding
origin/main + use_vertexai=True + no api_key	ServiceInitializationError: The API key is required when use_vertexai is False.	same	same
PR branch + use_vertexai=True + no api_key	initializes successfully	initializes successfully	initializes successfully
PR branch + use_vertexai=True + missing cloud_region	ServiceInitializationError: Region must be provided when use_vertexai is True.	same	same

So the rebased PR still fixes the original bug on all 3 Google AI service initializers while preserving the new cloud_region validation path.

[moonbox3]

Just need the CI/CD checks to pass and we'll get this merged. We will monitor.

[giulio-leone]

Rebased this branch onto the current main (0e2e6bd77) and reran the local gate twice from the rebased head 8ba149c9c with no code changes between passes.

Local validation (Pass 1 and Pass 2)

• UV_PYTHON=3.10 uv run --frozen --extra google --extra pandas ruff check semantic_kernel/connectors/ai/google/google_ai/services/google_ai_chat_completion.py semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_completion.py semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_embedding.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py
  • both passes: All checks passed!
• UV_PYTHON=3.10 uv run --frozen --extra google --extra pandas pytest tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py -q
  • both passes: 53 passed

Direct runtime proof against current main

Using the real source trees with:

• GOOGLE_AI_GEMINI_MODEL_ID=test-gemini-model-id
• GOOGLE_AI_EMBEDDING_MODEL_ID=test-embedding-model-id
• GOOGLE_AI_CLOUD_PROJECT_ID=test-project-id
• GOOGLE_AI_CLOUD_REGION=test-region
• no GOOGLE_AI_API_KEY

Results were identical across both proof passes:

Case	Chat	Text	Embedding
current main + use_vertexai=True + no api_key	ServiceInitializationError: The API key is required when use_vertexai is False.	same	same
rebased PR branch + use_vertexai=True + no api_key	initializes successfully	initializes successfully	initializes successfully
rebased PR branch + use_vertexai=True + missing cloud_region	ServiceInitializationError: Region must be provided when use_vertexai is True.	same	same

So after rebasing onto the latest main, this PR still fixes the original Vertex AI initialization bug on all three Google AI service initializers while preserving the fast-fail cloud_region validation path.

[giulio-leone]

Small status note on the refreshed head 8ba149c9c: local validation is in the comment above and GitHub has created fresh workflow runs for the new SHA, but the main workflows are currently showing action_required with zero jobs attached on this fork-based refresh.

If that is just the usual maintainer approval step for fork workflows, that looks like the remaining blocker on the new head. If not, I am happy to investigate further from my side.

[moonbox3]

@giulio-leone please check the integration test errors: https://github.com/microsoft/semantic-kernel/actions/runs/23415605290/job/68110402458#step:9:16509

[giulio-leone]

Follow-up for the integration-test feedback is now pushed on head 8fae8f7a1.

Root cause: the branch had added the new cloud_region fast-fail validation (which is still correct), but the Python integration selectors were still treating Vertex AI as "configured" when only GOOGLE_AI_CLOUD_PROJECT_ID + model ID were present. That meant the integration fixture tried to construct Vertex AI services even when GOOGLE_AI_CLOUD_REGION was missing, and the new validation then surfaced as broad setup failures across parametrized integration jobs.

Fix applied:

• python/tests/integration/completions/chat_completion_test_base.py
• python/tests/integration/completions/test_text_completion.py
• python/tests/integration/embeddings/test_embedding_service_base.py

All three now require GOOGLE_AI_CLOUD_REGION in the vertex_ai_setup gate before enabling Vertex AI test paths.

Validation from the refreshed head:

• Pass 1 and Pass 2, back-to-back with no code changes:
  • cd python && UV_PYTHON=3.10 uv run --frozen ruff check tests/integration/embeddings/test_embedding_service_base.py tests/integration/completions/chat_completion_test_base.py tests/integration/completions/test_text_completion.py
  • cd python && UV_PYTHON=3.10 uv run --frozen pytest tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py -q
• Clean-head sanity rerun on the committed HEAD: same commands passed again.

Direct proof of the exact integration trap (with GOOGLE_AI_CLOUD_PROJECT_ID + model IDs set but GOOGLE_AI_CLOUD_REGION unset):

• origin/main selector logic still reports Vertex AI as configured in all three integration files.
• refreshed branch selector logic now reports Vertex AI as not configured in all three integration files, so those paths skip instead of exploding during fixture setup.
• production validation is preserved on the refreshed branch: GoogleAIChatCompletion, GoogleAITextCompletion, and GoogleAITextEmbedding still all raise Region must be provided when use_vertexai is True. when actually instantiated without a region.

So this follow-up addresses the collaborator-reported integration failures without weakening the original use_vertexai=True / no-API-key fix or the new cloud_region validation.

[moonbox3]

@giulio-leone please make sure you have the pre-commit installed locally so all of these failures are caught before committing code to your branch and pushing upstream. There are ruff failures that need to be addressed.

[giulio-leone]

Addressed the latest pre-commit / ruff-format feedback on refreshed head f50d67670ff90c06bcebbb95d1d76aecf31db038.

This follow-up is formatter-only on the 3 integration selector files flagged by CI:

• python/tests/integration/completions/chat_completion_test_base.py
• python/tests/integration/completions/test_text_completion.py
• python/tests/integration/embeddings/test_embedding_service_base.py

What changed:

• normalized the multiline vertex_ai_setup = is_service_setup_for_testing([...]) lists to the exact ruff-format style that the pre-commit (3.10) job rewrote in CI

Local verification on the updated head:

• Pass 1 and Pass 2 before commit, back-to-back with no code changes:
  • cd python && UV_PYTHON=3.10 uv run --frozen pre-commit run --files tests/integration/completions/chat_completion_test_base.py tests/integration/completions/test_text_completion.py tests/integration/embeddings/test_embedding_service_base.py
  • cd python && UV_PYTHON=3.10 uv run --frozen pytest tests/unit/connectors/ai/google/google_ai/services/test_google_ai_chat_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_completion.py tests/unit/connectors/ai/google/google_ai/services/test_google_ai_text_embedding.py -q
• Clean-head sanity rerun after commit: same commands passed again.

So the branch now includes the exact formatter normalization CI requested, while the previously pushed integration-selector fix and the original Vertex AI behavior fix remain unchanged.

[giulio-leone]

PR refresh — 2025-07-10 🤖

Validated on current main 082e28e52 (already rebased, no changes needed).

Test results (double pass ✅)

• tests/unit/connectors/ai/google/google_ai/services/: 60/60 passed (×2)

[giulio-leone]

✅ Autonomous Validation — Pass 2/2 Clean

Branch: fix/issue-13483-google-ai-vertexai-api-key
HEAD: f50d67670f
Status: 0 commits behind origin/main — no rebase needed

Test Results (double-pass, no code changes between passes)

Pass	Tests	Result
1	60/60	✅
2	60/60	✅

Runner: python -m pytest tests/unit/connectors/ai/google/google_ai/services/ -v (Python 3.10)

Fix Verification

Google AI services skip api_key validation check when use_vertexai=True. Unit tests in test_google_ai_chat_completion.py, test_google_ai_text_completion.py, and test_google_ai_text_embedding.py all confirm correct behavior. Fix confirmed live in worktree; absent in upstream main.

Validated autonomously — double-pass strict protocol (identical runs, no edits between passes)