Gemini Enterprise Agent Platform (Vertex AI)

Page Assist has native support for Google Cloud's Gemini Enterprise Agent Platform (formerly Vertex AI) — no local proxy (like LiteLLM) is required. Page Assist authenticates directly from your browser by minting short-lived Google Cloud access tokens from your service-account key, and refreshing them automatically.

Naming

At Google Cloud Next 2026, Vertex AI was renamed to the Gemini Enterprise Agent Platform. The API endpoints, model IDs, and authentication are unchanged — only the product name differs. This page (and the in-app provider) is still listed under "Vertex AI" so it stays easy to find.

Vertex AI vs. Google AI

This is different from the Google AI provider, which connects to Google AI Studio with a simple API key. Use this provider if your Gemini access lives inside a Google Cloud project (billing, quotas, and IAM managed by Google Cloud).

Prerequisites

Before you start, make sure you have:

1. A Google Cloud project with the Vertex AI API enabled.
   • In the Google Cloud Console, go to APIs & Services → Enable APIs and Services and enable Vertex AI API.
2. A service account with permission to call Vertex AI (the Vertex AI User or Agent Platform User role, roles/aiplatform.user, is enough).
3. A service-account key in JSON format.
   • Console: IAM & Admin → Service Accounts → your account → Keys → Add key → Create new key → JSON.
   • This downloads a .json file that looks like:

     {
       "type": "service_account",
       "project_id": "my-gcp-project-123456",
       "private_key_id": "...",
       "private_key": "-----BEGIN PRIVATE KEY-----\n...\n-----END PRIVATE KEY-----\n",
       "client_email": "[email protected]",
       "...": "..."
     }

Configuration

1. Click on the Page Assist icon on the browser toolbar.

2. Click on the Settings icon.

3. Go to the OpenAI Compatible API tab.

4. Click on the Add Provider button.

5. Select Gemini Enterprise Agent Platform (Vertex AI) from the dropdown.

6. Fill in the form:

   Field	Description
   Provider Name	Any label you like (e.g. Vertex AI).
   Google Cloud Project ID	Your GCP project ID, e.g. my-gcp-project-123456.
   Location / Region	The region your models run in, e.g. us-central1. Choose global to use the global endpoint.
   Service Account JSON	Paste the full contents of the service-account .json key file.

7. Click on the Save button.

8. Page Assist will open the model list. Select the Gemini models you want to use and save.

Choosing models

After adding the provider you can pick from the available Gemini models (for example gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite, gemini-3.5-flash). You can also add a model manually from Manage Models → Add Custom Model if you need a specific model id.

INFO

Model ids for Vertex AI use the google/ publisher prefix on the OpenAI-compatible surface (for example google/gemini-2.5-flash). Page Assist handles this for you when you select a model from the list.

Model availability varies

Not every Gemini model is enabled on every project or in every region. In particular, some newer or "pro"-tier models (for example Gemini 3 Pro, id gemini-3-pro-preview) are in gated preview and will return a 404 NOT_FOUND — "your project does not have access" until your project is allowlisted, even though the model exists. This is a Google-side access gate, not a Page Assist issue.

To request access, open the Vertex AI Model Garden, find the model, and use Enable / Request access. Once granted, add the model with Manage Models → Add Custom Model (type the id, e.g. gemini-3-pro-preview).

If a model 404s, try another model, a different region (or global), or confirm access in the Model Garden. The Gemini 3 flash tier (gemini-3.5-flash, gemini-3-flash-preview, gemini-3.1-flash-lite) is generally available without special access.

How authentication works

• Your service-account JSON is stored locally in the extension and never sent anywhere except Google's official token endpoint (oauth2.googleapis.com).
• Page Assist signs a JWT with your service-account key (using the browser's Web Crypto API) and exchanges it for a short-lived access token, which is cached and refreshed automatically before it expires.
• Requests go directly to the regional Vertex AI endpoint: https://{LOCATION}-aiplatform.googleapis.com/.../endpoints/openapi (or the global endpoint when global is selected).

Quick alternative

Instead of a service-account JSON you can paste a raw access token (e.g. from gcloud auth print-access-token) into the Service Account JSON field. Note that these tokens expire after about an hour and will need to be replaced manually — the service-account JSON is recommended for everyday use.

Troubleshooting

• 401 / UNAUTHENTICATED — Check that the JSON is pasted in full and that the service account has the Vertex AI User role. If you pasted a raw access token, it may have expired.
• 403 / PERMISSION_DENIED — The Vertex AI API may not be enabled on the project, or the service account lacks permission.
• 404 / model not found — Confirm the model is available in the selected region, and that the model id is correct. Some models are region-specific; global covers the widest set.