Vertex AI: connect a Cloud project to your AI tools

Connect your AI tools to the Cloud project that pays for their model access.

Vertex connects AI tools and applications to models through a Google Cloud project. This route makes sense when your course or team already uses Cloud billing. An education award is optional. For class-project funding options, see AI credits for teaching.

If you only need a provider key for a desktop chat client, start with the API-key setup guide. Cherry Studio supports both routes; you do not need to move to Vertex to use it.

Google's current documentation calls this platform Gemini Enterprise Agent Platform, formerly Vertex AI. Some clients still label their connection “Vertex AI.” The service identifier remains aiplatform.googleapis.com. Follow the current Google setup documentation if console labels differ.

What connects to what

Three things must line up: the client's project ID, the identity making the request, and that project's billing account. Your Google browser login may differ from the identity a desktop client uses. Enabling the API and granting permission to call it are also separate steps.

Check your project

1. Open the Google Cloud console and select the project you intend to use.
2. Copy its project ID from the project information. Keep this beside the client settings you will configure later.
3. Open Billing and confirm the linked account is enabled. For a new project, an account administrator may need to link it. Follow Google's billing-status check if the connection is unclear. If you are using education credits, check their balance, expiry and eligible services in that account.
4. In APIs & Services, check whether aiplatform.googleapis.com is enabled for this project. An authorized administrator can enable it if needed.
5. Check that your intended model supports the client's location in Google's model and endpoint table. Use the provider-native model ID, not an alias copied from another gateway.

The global endpoint is not a data-residency guarantee. Check institutional requirements before choosing an endpoint or submitting research or student data.

Choose an identity

Choose the client before creating a credential. A browser sign-in, Application Default Credentials (ADC), an API key and a service-account JSON key are different routes. A client must support the route you configure.

Google's quickstart supports ADC or an API key and recommends ADC. That does not mean every desktop client supports both. If you are writing your own application, use Google's ADC instructions. Follow the client-specific guide for desktop applications.

Ask the project administrator to grant the permissions the actual calling identity needs. Google's predefined roles/aiplatform.user role is currently labeled Agent Platform User; older interfaces may say Vertex AI User. It is a platform role, not a Gemini-only permission or a spending limit. An administrator can consider a narrower custom role. Do not grant Owner or Editor just to make an error disappear. See Google's required-role instructions.

If a client requires a service-account key, confirm that your organization allows it. Prefer a supported route without downloaded keys when available. Keep credentials out of prompts, repositories, shared drives and screenshots. Do not distribute an instructor's key to a class. Google explains the risks and alternatives in its service-account key guidance.

Set cost controls

Alerts-only Cloud Billing budgets track spending; they do not automatically cap it. See Google's budget guidance to set alerts for the project.

Google also documents spend cap budgets in Preview for eligible services, including Agent Platform. They apply to one project and one service, with a monthly budget based on gross costs before credits. Enforcement is delayed: in-flight requests can finish, and persistent resources can keep accruing charges. Check availability and permissions in your account before relying on this feature. See spend cap limits and setup.

Start with one short request, then inspect usage and any credit attribution when billing data appears. A successful reply alone does not show whether credits paid for it.

Choose a client

Choose the interface you want to work in:

• Antigravity Desktop with Google Cloud to connect Google's desktop assistant through browser sign-in
• Cherry Studio through Vertex AI to use your Cloud project in Cherry's chat interface; this route imports a service-account key
• Google's API quickstart to give an application you build its own model access

An assistant's login does not automatically give software you build its own credentials or billing connection.

Find the step that failed

IAM changes can take time to propagate. Follow Google's propagation guidance and confirm the identity's role binding before adding broader permissions or repeating requests.

Keep only redacted diagnostics when asking for help: the client version, selected provider, model/location and error code. Never attach credentials or a service-account JSON file.

@misc{docs-implementation-ai-environment-setup-vertex-ai-setup-2026,
  author = {Xule Lin},
  title = {Vertex AI: connect a Cloud project to your AI tools},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/implementation/ai-environment-setup/vertex-ai-setup}},
  note = {ORCID: 0000-0001-7885-4194}
}