| name | integrating-wif-gcs-streamassist |
| description | Sets up, tests, and tears down a GCS Data Store linked to a Gemini Enterprise engine, with support for Workforce Identity Federation (WIF) access. Use when the user requests linking Cloud Storage buckets to Gemini Enterprise or testing WIF-based GCS search. |
Integrating WIF, GCS, and StreamAssist
This skill provides recipes and automations to deploy an unstructured Google Cloud Storage (GCS) Data Store, hook it up to a Gemini Enterprise engine, and run testing using Workforce Identity Federation (WIF) or standard Application Default Credentials (ADC).
When to use this skill
- Setting up a temporary or persistent GCS bucket for document search.
- Indexing PDFs from local storage (like
~/Downloads) to GCP.
- Connecting unstructured data stores to an existing Gemini Enterprise app.
- Querying the
streamAssist API with specific data store constraints.
- Cleaning up all resources after testing.
Workflow
To execute the deployment and verify it:
- Preparation: Ensure you are logged into GCP and have selected the target project (
vtxdemos).
- Deployment: Execute
setup.py using uv run. This creates the bucket, uploads a PDF, creates the Data Store, imports the document, and attaches it.
- Verification: Execute
test_search.py using uv run. Pass an optional query. By default it uses local ADC, but will auto-detect WIF token if /tmp/entra_token.txt is present.
- Teardown: Execute
teardown.py using uv run to delete the bucket, datastore, and detach it from the engine.
graph TD
A[Start] --> B[setup.py]
B --> B1[Create gs://bucket]
B1 --> B2[Upload PDF from Downloads]
B2 --> B3[Create Data Store]
B3 --> B4[Import GCS to Data Store]
B4 --> B5[Attach to Gemini Engine]
B5 --> C[test_search.py]
C --> D[teardown.py]
D --> D1[Detach Data Store]
D1 --> D2[Delete Data Store]
D2 --> D3[Delete gs://bucket]
D3 --> E[End]
Instructions
1. Running the Setup
Execute the following command in the workspace root:
uv run agy-recipes/ge_api_wif_gcs/scripts/setup.py
This script saves the generated resource names to last_setup_resources.json to allow clean teardown.
2. Testing the Search
To search the newly created GCS index using standard ADC credentials:
uv run agy-recipes/ge_api_wif_gcs/scripts/test_search.py "What are the financial highlights?"
To test using WIF, write your Microsoft Entra ID JWT token to /tmp/entra_token.txt first:
echo "YOUR_ENTRA_JWT_TOKEN" > /tmp/entra_token.txt
uv run agy-recipes/ge_api_wif_gcs/scripts/test_search.py "What are the financial highlights?"
3. Cleaning Up
To delete all resources created during setup and avoid any costs:
uv run agy-recipes/ge_api_wif_gcs/scripts/teardown.py
Resources
