Build and manage workloads

Use the portal to create a workload, prepare its building blocks, and bring a release through review. A workload represents an AI capability such as an assistant or classifier.

Before you start

Select the workspace where the workload will belong. You need permission to create and manage the resources for your task. Your organization may require a different person to approve a release.

Follow the workload journey

1. Create a workload: describe its purpose, interface, governance needs, and sandbox.
2. Manage assets and versions: prepare the prompts, flows, model settings, and policies it needs.
3. Create or refine its workflow, then test and evaluate the saved candidate.
4. Prepare a release: bind the chosen versions, validate, and submit for review.
5. Review a release: inspect the evidence and record a decision with an authorized reviewer.
6. Activate the approved release with an authorized operator and check workload readiness.
7. Investigate runs and human decisions after dispatch.

Registration, approval, and activation are separate milestones. A registered workload may still need an active release, a service lane, or a required access or human decision before it can serve requests.

Prepare access and integrations

• Understand permissions.
• Understand namespaces and workspaces.
• Configure AI resources and the workspace assistant.
• Configure connections and secrets.

If you are blocked

Read the readiness reasons on the workload or release and resolve the named requirement. Ask your workspace administrator for access issues and the workload owner for configuration questions. See Troubleshooting before repeating an action with an uncertain result.

API, SDK, CLI, and advanced configuration are in Technical documentation.