Agent Platform in Cloud de Confiance versus Google Cloud

Gemini Enterprise Agent Platform is a platform for building, deploying, and managing generative AI applications. This page describes the differences between the Cloud de Confiance and Google Cloud versions of Agent Platform.

For more detailed information about Agent Platform, see the Agent Platform overview and the rest of the Agent Platform documentation.

Key differences

There are some differences between the Cloud de Confiance version of Agent Platform and the Google Cloud version. Some notable differences include the following:

  • Only Model Garden, Model Registry on Gemini Enterprise Agent Platform, and Online inference (predictions) are available in Cloud de Confiance. Other components and features of Agent Platform are unavailable.
  • The Agent Platform SDK (Vertex AI SDK) and Google Gen AI SDK are unavailable. Supported interfaces are the Cloud de Confiance console, the Google Cloud CLI (gcloud CLI), and the REST API.
  • Open first-party and third-party models (such as Gemma and Mistral) are available. Gemini and other proprietary models are unavailable.
  • Only self-deploy options are available in Cloud de Confiance.

A more detailed list of differences is provided in the rest of this section. If you are already familiar with Google Cloud, we recommend that you review these differences carefully, particularly before designing an application to run on Cloud de Confiance. We also recommend reviewing the general differences between Cloud de Confiance and Google Cloud.

If you would like to use a particular Agent Platform feature that isn't currently available in Cloud de Confiance, contact Cloud de Confiance support. To be notified when new features roll out in Cloud de Confiance, subscribe to the release notes. Unless otherwise specified, features that are in preview are not available in Cloud de Confiance.

Workflows and tools

Feature support differences
SDK and client interfaces The Agent Platform SDK (Vertex AI SDK) and Google Gen AI SDK are unavailable in Cloud de Confiance. Supported interfaces are the Cloud de Confiance console, the gcloud CLI (gcloud CLI), and the REST API.
Unavailable components The following components of Agent Platform are unavailable in Cloud de Confiance: all components other than Model Garden, Model Registry, and Online inference.

Model Garden

Functionality and options differences
Features not available in the console interface The following areas are not available in the main console interface:
  • Explore Generative AI
  • Fast Deployment
  • Deployment with custom weights
  • Foundational Models section (including Gemini models)
  • Fine-tunable Models section
Model availability Gemini models and third-party proprietary models are unavailable. Gemma and select third-party open-source models are available.
Model deployment and actions The following actions are unavailable:
  • Model fine-tuning
  • Opening a model notebook
  • Model evaluation
  • "Try it Out" (Agent Studio feature)
  • One-click deployment of external models directly from Hugging Face Hub
Model deployment options Only self-deploy options are available. Model-as-a-Service is unavailable.
Deployment targets Deploying models from Model Garden to Google Kubernetes Engine (GKE) is unavailable (only deployment to an Agent Platform endpoint is supported).

Model Registry

Functionality differences
Unavailable registry actions The following actions are unavailable in Model Registry in Cloud de Confiance:
  • Create a new model
  • Train a new model version
  • Register BigQuery ML models in Model Registry
  • Batch inference
  • Evaluate
  • Data lineage
  • Deploy a model for batch inference
  • Export model versions (ExportModel) to Cloud Storage
Model details page and actions Testing models is unavailable on the model details page.

Online inference (Predictions)

Functionality differences
Inference features Only synchronous Predict and RawPredict prediction methods are available. The following inference features and prediction methods are unavailable in Cloud de Confiance:
  • Streaming prediction methods (StreamRawPredict, ServerStreamingPredict, StreamingPredict, and StreamGenerateContent)
  • Specialized prediction methods (ChatCompletions, DirectPredict, CountTokens, and EmbedContent)
  • Explanation requests (Explain)
  • Invoke requests
  • Batch prediction (batch prediction API endpoints are omitted or disabled)
Endpoint access Only standard-access regional endpoints are available (Private Service Connect and private peering endpoints are unavailable).

Integrations

Integration differences
BigQuery ML integration You can't register BigQuery ML models in Model Registry in Cloud de Confiance.

API surface

API surface differences
API endpoint restrictions Only allowed control plane methods (for endpoints, models, and deployments) and data plane methods (for serving predictions) are available. Specifically, only synchronous Predict and RawPredict prediction methods are available. Streaming prediction methods (StreamRawPredict, ServerStreamingPredict, StreamingPredict, and StreamGenerateContent), specialized prediction methods (ChatCompletions, DirectPredict, CountTokens, and EmbedContent), and other methods (such as Explain, GenerateContent, and ExportModel) are unavailable.

The following information might also affect how you use and design for Agent Platform in Cloud de Confiance by S3NS. These guides include general information about working in Cloud de Confiance, including documentation, security and access control, billing, tooling, and service usage.

For details about other services and features in Cloud de Confiance and their differences from their Google Cloud counterparts, see the product list.