Skip to main content

GCP Ai Platform Endpoint

A Vertex AI (formerly AI Platform) Endpoint is a regional resource that serves as an entry-point for online prediction requests in Google Cloud. One or more trained Models can be deployed to an Endpoint, after which client applications invoke the Endpoint’s HTTPS URL (or Private Service Connect address) to obtain real-time predictions. The resource stores configuration such as traffic splitting between models, logging settings, encryption settings and the VPC network to be used for private access.
Official documentation: https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.endpoints

Supported Methods​

  • GET: Get a gcp-ai-platform-endpoint by its "name"
  • LIST: List all gcp-ai-platform-endpoint
  • SEARCH

gcp-ai-platform-model​

An Endpoint may contain one or many deployedModel blocks, each of which references a separate Model resource. Overmind links the Endpoint to every Model that is currently deployed or that has traffic allocated to it.

gcp-ai-platform-model-deployment-monitoring-job​

If model-deployment monitoring has been enabled, the monitoring job resource records statistics and drift detection for a specific Endpoint. Overmind links the Endpoint to all monitoring jobs that target it.

gcp-big-query-table​

Prediction logging and monitoring can be configured to write request/response data into BigQuery tables. Those tables are therefore linked to the Endpoint that produced the records.

gcp-cloud-kms-crypto-key​

Endpoints can be created with a Customer-Managed Encryption Key (CMEK) via the encryptionSpec.kmsKeyName field. Overmind links the Endpoint to the specific Cloud KMS CryptoKey it uses for at-rest encryption.

gcp-compute-network​

When an Endpoint is set up for private predictions, it must specify a VPC network (network field) that will be used for Private Service Connect. This creates a relationship between the Endpoint and the referenced Compute Network.