canonical: https://jentic.com/apis/azure.com/azure-machine-learning-compute-management-client

# Microsoft Azure Machine Learning Compute Management Client

Jentic publishes the only available OpenAPI specification for Machine Learning Compute Management Client, keeping it validated and agent-ready. This API manages the legacy Microsoft.MachineLearningCompute provider - the operationalization clusters that hosted scoring endpoints for Azure ML models in the original AML Workbench era. It provisions clusters, retrieves their keys, refreshes system service updates, and inspects update availability. It does not train models or run scoring requests.

## For AI agents

Provision and operate legacy Azure ML operationalization clusters used to host model scoring endpoints. Use for cluster lifecycle and system service updates, not for training models or scoring inference requests.

## Scope

Does not train models, run scoring requests, or manage modern Azure ML workspaces - use only for legacy Microsoft.MachineLearningCompute operationalization cluster lifecycle and system service updates.

## Capabilities

- Provision and delete operationalization clusters that host Azure ML scoring endpoints
- Retrieve cluster access keys for downstream deployment tooling
- Check whether system service updates are available for an existing cluster
- Apply system service updates to keep clusters on supported infrastructure versions
- List operationalization clusters across a subscription or resource group

## Use cases

### Legacy AML Scoring Cluster Operations

Teams still running Azure ML Workbench-era operationalization clusters need to keep them patched and reachable. The CheckSystemServicesUpdatesAvailable and UpdateSystemServices operations let operators detect when an underlying Kubernetes or Azure Container Service version has shifted and apply the upgrade through Azure Resource Manager. This avoids a full cluster rebuild for security and version-currency updates.

Example prompt: Call CheckSystemServicesUpdatesAvailable on the 'aml-scoring-prod' cluster, and if updatesAvailable is true, call UpdateSystemServices and poll until the operation completes.

### Cluster Inventory and Key Rotation

Security teams auditing Azure ML scoring infrastructure list every operationalization cluster across the subscription, then call ListKeys per cluster to inventory which downstream tooling holds active credentials. Combined with periodic key regeneration, this supports compliance reviews of model-serving credentials.

Example prompt: List all OperationalizationClusters in the subscription, then for each cluster call ListKeys and produce a CSV of cluster name, resource group, and storage account key fingerprint.

### Cluster Decommission Workflow

Organisations migrating from Azure ML Workbench scoring to Azure Machine Learning workspaces need to decommission the legacy operationalization clusters in a controlled order. Listing clusters, validating no active deployments, then calling Delete drains the resource cleanly. The DELETE on /operationalizationClusters/{clusterName} is asynchronous and returns 202 Accepted with an Azure-AsyncOperation URL to poll.

Example prompt: List operationalization clusters tagged decom=true, then DELETE each cluster and poll its Azure-AsyncOperation URL until the operation reports Succeeded.

### AI Agent Cluster Maintenance

Platform agents running scheduled maintenance use Jentic to find the right operationalization-cluster operation without bundling the legacy AML SDK. The agent searches an intent like 'check azure ml cluster system updates', loads the operation schema, and executes the call against management.azure.com with a token retrieved from your Jentic One instance.

Example prompt: Use Jentic to search 'check azure ml operationalization cluster system updates', load the schema, and execute against the target cluster, returning the updatesAvailable flag and the recommended update list.

## Key endpoints

| Method | Path | Description |
| --- | --- | --- |
| PUT | /subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.MachineLearningCompute/operationalizationClusters/{clusterName} | Create or update an operationalization cluster |
| GET | /subscriptions/{subscriptionId}/providers/Microsoft.MachineLearningCompute/operationalizationClusters | List operationalization clusters in a subscription |
| DELETE | /subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.MachineLearningCompute/operationalizationClusters/{clusterName} | Delete an operationalization cluster |
| POST | /subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.MachineLearningCompute/operationalizationClusters/{clusterName}/checkSystemServicesUpdatesAvailable | Check whether system service updates are available |
| POST | /subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.MachineLearningCompute/operationalizationClusters/{clusterName}/updateSystemServices | Apply system service updates to a cluster |

## Key resources

- **OperationalizationClusters** — Legacy Azure ML scoring clusters: create, list, get, update, delete, list keys, check and apply system service updates.
- **MachineLearningComputeOperations** — Operations catalog exposed by the Microsoft.MachineLearningCompute resource provider.

## Why Jentic

- **Setup:** Wiring the legacy Microsoft.MachineLearningCompute API by hand means registering an Azure AD app, negotiating OAuth 2.0 tokens for management.azure.com, and threading subscription and resource-group scope through every operationalization-cluster call. Through Jentic you install once, import Machine Learning Compute Management Client from the API Directory, store the Azure AD credential once, and your agent calls it.
- **Permission scoping:** This API puts the operationalization cluster in the URL path (/operationalizationClusters/{clusterName}), so a rule can pin your agent to one cluster: it can list clusters and check for available system-service updates. You choose the operations it may call, so creating, deleting, or updating a cluster's system services is not included unless you add it.
- **Credential handling:** Your Azure AD credential for the MachineLearningCompute control plane is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
- **Discovery method:** Agents search Jentic by intent such as 'list ML operationalization clusters' or 'check for cluster system-service updates', and Jentic returns the matching operation with its input schema so the agent calls the right endpoint without browsing the reference docs.

## Related APIs

- **Machine Learning Workspaces Management Client** — Manages the modern Azure ML workspace plane that supersedes operationalization clusters.
- **Azure Machine Learning Model Management Service** — The legacy data-plane model management surface for the same operationalization stack.
- **Azure Machine Learning Datastore Management Client** — Manage the datastores that ML training and scoring jobs consume.

## FAQ

### Why is there no official OpenAPI spec for Machine Learning Compute Management Client?

Microsoft Azure does not publish an OpenAPI specification. Jentic generates and maintains this spec so that AI agents and developers can call Machine Learning Compute Management Client via structured tooling. It is validated against the live API and kept up to date. Get started with Jentic One, the self-hosted execution layer.

### What authentication does the Machine Learning Compute Management Client use?

AAD OAuth 2.0 bearer tokens scoped to https://management.azure.com/. With Jentic, the AAD client credentials are stored encrypted in the vault and a short-lived bearer token is injected at call time.

### Can I score a model with this API?

No. Scoring is a data-plane call against the cluster's exposed scoring endpoint, not against management.azure.com. This client only provisions, lists, updates, and deletes the cluster resource and exposes its access keys.

### What are the rate limits for the Machine Learning Compute Management Client?

Standard Azure Resource Manager throttling: typically 12,000 reads and 1,200 writes per hour per subscription per region. Cluster create, update, and delete operations are asynchronous and return 202 Accepted with an Azure-AsyncOperation URL to poll for completion.

### How do I check for cluster system updates through Jentic?

Run pip install jentic, then search Jentic for 'check azure ml operationalization cluster system updates'. Jentic returns the OperationalizationClusters_CheckSystemServicesUpdatesAvailable operation; load it, supply the cluster resource ID, and execute the POST call.

### Should I use this for new Azure ML projects?

No - this is the legacy AML Workbench operationalization plane. New Azure Machine Learning workloads should use the Microsoft.MachineLearningServices provider (workspaces, online endpoints, batch endpoints). Use this API only when maintaining existing operationalization clusters.

### Can I limit what my agent is allowed to do with the Machine Learning Compute Management Client?

Yes. Because this API carries the operationalization cluster name in the URL path (/operationalizationClusters/{clusterName}), your self-hosted Jentic One instance lets you write rules that pin the agent to a single cluster and to specific operations, such as listing clusters and checking whether system-service updates are available. You decide which operations the agent may call, so create, delete, and update-system-services calls stay off limits unless you explicitly add them. The Azure AD credential you store stays with your own Jentic One instance and is injected only when an allowed operation runs.