GCP Vertex AI

​

Overview

​

Step 1: Prepare Your GCP Environment

​

1.1 Create or Use a GCP Project

• Sign in to the GCP Console:
  Google Cloud Console
• Select or Create a Project:
  Use an existing project or create a new one dedicated to Vertex AI.

​

1.2 Set Up IAM Permissions and Service Accounts

• Assign Required Roles:
  • Grant your user (or service account) the Vertex AI User role (roles/aiplatform.user)
  • For service accounts, also attach the Vertex AI Service Agent role (roles/aiplatform.serviceAgent) to enable certain operations
  • Consider additional predefined roles as needed:
    • Vertex AI Platform Express Admin
    • Vertex AI Platform Express User
    • Vertex AI Migration Service User
• Cross-Project Resource Access:
  • For BigQuery tables in different projects, assign the BigQuery Data Viewer role
  • For Cloud Storage buckets in different projects, assign the Storage Object Viewer role
  • For external data sources, refer to the GCP Vertex AI Access Control documentation

​

Step 2: Verify Regional and Model Access

​

2.1 Choose and Confirm a Region

• us-east5 (Columbus, Ohio)
• us-central1 (Iowa)
• europe-west1 (Belgium)
• europe-west4 (Netherlands)
• asia-southeast1 (Singapore)
• global (Global)

​

2.2 Enable the Claude 3.5 Sonnet v2 Model

• Open Vertex AI Model Garden:
  In the Cloud Console, navigate to Vertex AI → Model Garden
• Enable Claude 3.5 Sonnet v2:
  Locate the model card for Claude 3.5 Sonnet v2 and click Enable

​

Step 3: Configure the Cline VS Code Extension

​

3.1 Install and Open Cline

• Download VS Code:
  Download Visual Studio Code
• Install the Cline Extension:
  • Open VS Code
  • Navigate to the Extensions Marketplace (Ctrl+Shift+X or Cmd+Shift+X)
  • Search for Cline and install the extension

​

3.2 Configure Cline Settings

• Open Cline Settings:
  Click the settings ⚙️ icon within the Cline extension
• Set API Provider:
  Choose GCP Vertex AI from the API Provider dropdown
• Enter Your Google Cloud Project ID:
  Provide the project ID you set up earlier
• Select the Region:
  Choose one of the supported regions (e.g., us-east5)
• Select the Model:
  From the available list, choose Claude 3.5 Sonnet v2
• Save and Test:
  Save your settings and test by sending a simple prompt (e.g., “Generate a Python function to check if a number is prime.”)

​

Step 4: Authentication and Credentials Setup

​

Option A: Using Your Google Account (User Credentials)

1. Install the Google Cloud CLI:
   Follow the installation guide
2. Initialize and Authenticate:
   Copy

   Ask AI

   gcloud init
   gcloud auth application-default login
   

   • This sets up Application Default Credentials (ADC) using your Google account
3. Restart VS Code:
   Ensure VS Code is restarted so that the Cline extension picks up the new credentials

​

Option B: Using a Service Account (JSON Key)

1. Create a Service Account:
   • In the GCP Console, navigate to IAM & Admin > Service Accounts
   • Create a new service account (e.g., “vertex-ai-client”)
2. Assign Roles:
   • Attach Vertex AI User (roles/aiplatform.user)
   • Attach Vertex AI Service Agent (roles/aiplatform.serviceAgent)
   • Optionally, add other roles as required
3. Generate a JSON Key:
   • In the Service Accounts section, manage keys for your service account and download the JSON key
4. Set the Environment Variable:
   Copy

   Ask AI

   export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/service-account-key.json"
   

   • This instructs Google Cloud client libraries (and Cline) to use this key
5. Restart VS Code:
   Launch VS Code from a terminal where the GOOGLE_APPLICATION_CREDENTIALS variable is set

​

Step 5: Security, Monitoring, and Best Practices

​

5.1 Enforce Least Privilege

• Principle of Least Privilege:
  Only grant the minimum necessary permissions. Custom roles can offer finer control compared to broad predefined roles
• Best Practices:
  Refer to GCP IAM Best Practices

​

5.2 Manage Resource Access

• Project vs. Resource-Level Access:
  Access can be managed at both levels. Note that resource-level permissions (e.g., for BigQuery or Cloud Storage) add to, but do not override, project-level policies

​

5.3 Monitor Usage and Quotas

• Model Observability Dashboard:
  • In the Vertex AI Console, navigate to the Model Observability dashboard
  • Monitor metrics such as request throughput, latency, and error rates (including 429 quota errors)
• Quota Management:
  • If you encounter 429 errors, check the IAM & Admin > Quotas page
  • Request a quota increase if necessary
    Learn more about GCP Vertex AI Quotas

​

5.4 Service Agents and Cross-Project Considerations

• Service Agents:
  Be aware of the different service agents:
  • Vertex AI Service Agent
  • Vertex AI RAG Data Service Agent
  • Vertex AI Custom Code Service Agent
  • Vertex AI Extension Service Agent
• Cross-Project Access:
  For resources in other projects (e.g., BigQuery, Cloud Storage), ensure that the appropriate roles (BigQuery Data Viewer, Storage Object Viewer) are assigned

​

Conclusion

• Prepare Your GCP Environment:
  Create or use a project, configure IAM with least privilege, and ensure necessary roles (including the Vertex AI Service Agent role) are attached
• Verify Regional and Model Access:
  Confirm that your chosen region supports Claude 3.5 Sonnet v2 and that the model is enabled
• Configure Cline in VS Code:
  Install Cline, enter your project ID, select the appropriate region, and choose the model
• Set Up Authentication:
  Use either user credentials (via gcloud auth application-default login) or a service account with a JSON key
• Implement Security and Monitoring:
  Adhere to best practices for IAM, manage resource access carefully, and monitor usage with the Model Observability dashboard