Onaro™ONARO

Documentation / Integrations / Google Vertex AI

Google Vertex AI connector

Connect Google Cloud BigQuery billing export to pull Vertex AI spend.

AvailableMethod: API

Overview

Connect Google Cloud BigQuery billing export to pull Vertex AI spend.

Data pulled

  • Daily net spend by SKU from the detailed billing export table
  • Filtered to service.description = Vertex AI
  • Token counts are not in the export (cost only)

What gets attributed

Meridian reads your Cloud Billing BigQuery export filtered to the Vertex AI service and stores one row per day for each model name derived from the SKU description, net of credits. It does not read project ids, labels or resource names from the export, so rows carry no team or agent information.

Every Vertex AI row arrives unattributed in the attribution inbox (Agents & allocation → Unattributed), where you can assign it or create a rule that matches on vendor or model.

Authentication & credential storage

Meridian stores the following fields for this connector (values encrypted at rest):

  • service_account_json
  • project_id
  • dataset_id
  • billing_account_id
  • Credentials are encrypted with AES-256-GCM using a KMS-wrapped data encryption key (DEK) and stored in vendor_connections.credentials_encrypted.
  • Meridian never returns the plaintext key after save. Revocation is disconnect: deactivate or delete the vendor connection — sync stops and the encrypted blob is no longer used.
  • Grant the service account roles/bigquery.jobUser and roles/bigquery.dataViewer on the billing-export dataset.

Sync schedule & behavior

Billing-export lag is typically 1–6 hours. Syncs automatically every 6 hours; manual sync anytime. There is no per-connection frequency setting.

Setup steps

  1. Enable detailed Cloud Billing export to BigQuery.
  2. Create a service account with BigQuery job and data viewer roles; download the JSON key.
  3. Connect GCP Vertex AI in Meridian with project, dataset, billing account id, and service account JSON.

Troubleshooting

  • Table not found: confirm the export table name matches gcp_billing_export_resource_v1_{billing_account}.
  • Permission denied: add bigquery.jobUser and bigquery.dataViewer.

Related

Onaro Meridian is FinOps for agentic AI: the system of record that attributes, controls and books what AI agents spend.