Gemini Vertex AI Setup with Workcell

Use this pattern when Gemini should authenticate through Vertex instead of a plain Gemini API key.

1. Create the Gemini env file

GOOGLE_GENAI_USE_VERTEXAI=true
GOOGLE_CLOUD_PROJECT=my-project
GOOGLE_CLOUD_LOCATION=us-central1

If you need ADC, export a reviewed copy. Keep it as a separate credential file in Workcell operator state.

2. Create the injection policy

version = 1

[credentials]
gemini_env = "/Users/example/.config/workcell/gemini-vertex.env"
gcloud_adc = "/Users/example/.config/workcell/gcloud-adc.json"

3. What Workcell does with this

• seeds ~/.gemini/.env inside the session
• treats ADC as a supplemental input rather than a standalone auth mode
• derives the matching regional aiplatform.googleapis.com allowlist entry for strict-mode launches when a location is present
• reports the gcloud_adc path as vertex-supplement with manual bootstrap support in workcell auth status, workcell --auth-status, and workcell why

4. Launch

workcell --prepare --agent gemini --workspace /path/to/repo

Related docs

• docs/injection-policy.md
• docs/provider-bootstrap-matrix.md
• docs/examples/quickstart-gemini.md
• docs/requirements-validation.md
• docs/validation-scenarios.md