Cloudera Docs

Deploying workflows as model endpoints

The Cloudera AI Agent Studio's deployment system enables you to transform AI workflows into production-ready endpoints. When deployed, each workflow operates as an independent service, leveraging both Cloudera AI Workbench models and a Cloudera AI Workbench application. For more information see, Models overview Models overview and Analytical Applications Analytical Applications.

A deployed workflow consists of two main components:
  1. A Cloudera AI Workbench Model that functions as the workflow engine to execute tasks.
  2. A Cloudera AI Workbench Application that provides a user interface for interacting with the workflow.

Workflows are fundamentally asynchronous. When a workflow start request is sent to a deployment, a trace ID is immediately returned to track the workflow’s status.

Workflow deployment logs and associated telemetry are sent to Agent Studio's Agent Operations and Metrics system. The typical workflow execution proceeds as follows:

  1. The workflow can be initiated either by starting it from the deployed workflow Application or by sending a kickoff request directly to the model endpoint.

  2. The workflow runs asynchronously within the model deployment, simultaneously streaming events and logs to the Operations and Metrics server.

  3. The workflow status can be checked and monitored for completion by polling the /events endpoint on the Operations & Metrics server.

For more information, see Monitoring feature in Agent StudioMonitoring feature in Agent Studio.