Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Microsoft Foundry governance, previously known as Foundry Control Plane, is the centralized management experience for enterprise AI fleets.
Microsoft Foundry provides built-in governance for AI agents, models, and tools through agent identity, agent access policies, cost management, and safety guardrails. Platform teams use supported policy controls to establish organizational boundaries, and engineers build within those boundaries.
Foundry governance combines built-in controls, centralized management, and optional integrations. Through Manage in the Foundry portal, administrators, platform teams, and developers can understand their AI assets, monitor operations, investigate costs, and assess policies and protections.
Extend this built-in foundation where needed with AI Gateway for centralized traffic controls, Microsoft Agent 365 for enterprise agent management, Microsoft Purview for data security and governance, and Microsoft Defender for threat protection and response through supported integrations.
These integrations extend Foundry's governance foundation. They aren't prerequisites for every native control, and they aren't one interchangeable control plane. Available capabilities depend on the workload, permissions, configuration, and supported integration.
Prerequisites and permissions
To use Foundry governance capabilities, you need:
-
An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription.
A Foundry project. If you don't have one, create a project.
- Access to the Foundry resources and projects that you want to review.
- Appropriate Azure role-based access control permissions for the selected scope and underlying services.
Individual capabilities can require additional permissions or configuration:
- Fleet monitoring: Access to the relevant Application Insights resources and log data. See Configure fleet monitoring.
- Costs: Access to billing and forecasts at the required scope, which can be broader than a project. See Permissions to view costs.
- Policies: Permission to read relevant assignments and compliance results. Assignment changes and exemptions require separate permissions in Azure Policy.
- Security and data governance: Configured coverage and access for Microsoft Defender for Cloud and Microsoft Purview.
An AI gateway isn't required for every Foundry governance capability. Agent 365, Purview, and Defender also aren't prerequisites for every native identity, access, cost, or safety control.
Note
This capability is available only in the Foundry (new) portal. Look for
in the portal banner to confirm you're using Foundry (new).
When to use Foundry governance
Use Foundry governance capabilities when your organization needs to:
- Establish supported governance boundaries for agent access, costs, and safety, and review their application across projects.
- Understand supported agents, models, and tools across teams.
- Identify assets contributing to errors, latency, increased consumption, or spending.
- Review policy requirements, assignment scopes, and noncompliant assets.
- Investigate security recommendations and data protection gaps from configured integrations.
- Move from fleet-level oversight to project-level investigation and the service responsible for a control.
Coverage differs across asset types and platforms. An external agent doesn't necessarily have the same inventory, telemetry, or management support as a Foundry agent. See Supported agent platforms.
Built-in governance, extended where needed
Foundry's built-in governance spans four areas. Together, they help your organization establish who an agent is, what it can access, how its usage is managed, and which safety controls apply.
| Area | Built-in capabilities | Learn more |
|---|---|---|
| Agent identity | Represent agents through Microsoft Entra agent identities and blueprints. Use supported authentication and on-behalf-of-user (OBO) delegation flows for downstream access. | Agent identity concepts and user delegation for tools. |
| Agent access | Apply policies for model deployments, agent kinds and hosted-agent container images, approved tools, and network egress. Coverage and enforcement vary by policy and agent type. | Model deployment policies, agent deployment policies, tool governance policies and controls, and hosted-agent network egress policies. |
| Cost management and FinOps | Monitor costs and consumption, manage model quotas, and configure cost controls (preview) with supported spending or token budgets and alert or block actions. | View subscription costs, manage quotas, and cost controls overview. |
| Safety and responsible AI | Configure guardrails and supported runtime controls for risks in inputs, outputs, and agent interactions. Applicability differs between models, agent types, and intervention points. | Guardrails and controls overview. |
Built-in doesn't mean that every control is enabled automatically or applies identically to every runtime. Some capabilities are in preview. Configure the controls required for your workload and review their supported scope and availability in the linked documentation.
Understand where enforcement happens
Azure Policy and enforcement in the relevant services form part of the governance foundation. Manage provides visibility into the estate, with Compliance showing policy assignments and available compliance results. The management experience isn't a substitute for enforcement.
For example:
- Model deployment policies govern approved models and deployment eligibility.
- Agent deployment policies restrict agent kinds and container images for applicable hosted agents.
- Native tool governance uses Azure Policy for supported configuration checks and separate project controls for runtime tool access. Assigning a policy doesn't automatically configure the project runtime controls.
- Hosted-agent egress governance checks supported egress configurations. Runtime enforcement requires the appropriate guardrail configuration.
Not every control uses the same policy mechanism or evaluation point. Review configuration-time checks and runtime restrictions separately. A compliant configuration doesn't establish that every runtime action is permitted or that all governance requirements have been assessed.
Extend the foundation through integrations
Connect additional services when your requirements extend beyond the native capabilities needed for your workload.
| Integration | What it adds | Scope to understand | Learn more |
|---|---|---|---|
| AI Gateway | Centralized traffic controls for supported model and tool endpoints, including routing, rate or token limits, and monitoring. | Controls apply to traffic routed through the configured gateway. Native Foundry governance doesn't universally require a gateway. | Configure AI Gateway for Foundry. |
| Microsoft Agent 365 | Enterprise agent inventory, identity-based governance, observability, and lifecycle management across supported platforms. | Agent support, telemetry, administrative actions, and enablement requirements vary by integration. | Agent 365 integration with Foundry. |
| Microsoft Purview | Data security and governance capabilities, including audit, data protection insights, and supported data loss prevention scenarios. | Coverage depends on the scenario, authentication, configuration, and licensing. Audit visibility isn't the same as policy enforcement. | Purview data security and compliance for Foundry. |
| Microsoft Defender | Security posture, threat protection, and investigation and response capabilities through supported integrations. | Defender for Cloud recommendations in Foundry and threat-protection or response workflows have distinct configuration and coverage requirements. | Set up Defender for Cloud and enable threat protection for AI services. |
These services complement one another. Don't assume universal policy synchronization, identical coverage across runtimes, or automatic correlation of all findings.
Explore the Manage experience
In the Foundry portal, select Manage to access centralized fleet management and administrative capabilities.
| Experience | What it helps you do |
|---|---|
| Fleet | Explore inventory, activity, token consumption, cost, and policy-violation highlights. |
| Cost | Analyze spending and forecasts, and manage applicable cost controls. |
| Operations | Review invocations, errors, latency, token usage, and available agent quality information. |
| Compliance | Review policies, asset compliance, Defender for Cloud recommendations, and Purview insights. |
| Quota | Review applicable model quota and access quota-management capabilities. |
| Gateways | Access gateway-management capabilities. |
Resource details and Project details provide entry points for resource and project administration.
Scope controls vary by page. Fleet, Cost, and Operations support selecting a subscription and narrowing the view by resource group or project. Check the selected scope and time range as you move between pages. Manage isn't limited to the currently selected project.
Understand your AI fleet
Fleet is the starting point for understanding assets and activity in your selected scope. Summary cards show projects, active agents and models, agent runs, token consumption, cost, and assets with policy violations.
Operational and governance highlights identify assets with high error incidents, high agent-run latency, cost increases, and policy violations. Use the inventory and contextual links to continue into Operations, Cost, or the relevant policy view.
For example, identify an agent with frequent errors, investigate its runtime behavior in Operations, and review its applicable requirements in Compliance. Compare resource identities and scopes rather than assuming that the findings share a cause.
An unavailable metric doesn't mean zero activity or no violations. For the page's data sources, filters, and investigation paths, see Monitor your agent fleet.
Understand and manage AI costs
Use Cost to investigate spending patterns and identify the projects, agents, and models contributing to costs within the selected subscription.
The Overview tab shows costs from Azure Cost Management, including the agent and model breakdowns. Its Estimated cost label reflects that current-month costs can change before the invoice is final. Cost data arrives later than operational usage metrics; don't treat token counts as billing records or assume that both views update together.
Available forecasts and comparisons depend on permissions and data. A missing forecast doesn't mean zero spending, and missing previous-period data can make comparisons unavailable.
The Cost Control tab provides access to Foundry cost controls. These controls maintain separate running totals for spending or tokens and apply the threshold actions you configure. They can alert or block new applicable requests when a threshold is reached.
A budget or dashboard summary alone doesn't guarantee that spending stops. Blocking requires a configured Block usage action. The request that crosses a threshold can still complete, and cost-control totals aren't the same as reconciled billing data.
For detailed guidance, see View subscription costs, cost controls overview, and create and manage cost controls.
Monitor operations across your fleet
Operations provides a cross-project view of activity and reliability for supported agents and models. Review agent runs, model requests, tool calls, and total tokens. Switch between Invocations, Tokens, Errors, and Latency to examine trends and identify assets that need investigation.
Available agent quality information depends on configured evaluations and results. An unavailable quality value isn't a low score.
Operational metrics help you decide where to investigate. They don't automatically establish a root cause or guarantee a matching trace for every event. Continue into project-level monitoring for available metrics, evaluations, and traces.
See Monitor agent and model operations and Monitor agents with the Agent Monitoring Dashboard.
Monitor compliance and security
Manage > Compliance brings together complementary governance perspectives.
| Tab | Focus |
|---|---|
| Policy | Policy and initiative assignments, scopes, compliance results, violating assets, and enforcement-mode information. |
| Asset | Available policy evaluations and impact for agents, models, and toolboxes. |
| Security posture | Defender for Cloud recommendations about security configuration risks and remediation. |
| Data security and governance | Purview coverage indicators, data protection gaps, and prioritized investigation actions. |
Start with a policy to understand its scope and impact, or start with an asset to review applicable requirements. Then continue into Azure Policy, Purview, Defender, or the relevant Foundry configuration experience.
Compliance centralizes visibility and investigation. It doesn't author or enforce every control itself. A compliance result isn't necessarily runtime blocking, and a Purview investigation lead isn't proof of malicious activity.
For detailed workflows and integration requirements, see Monitor AI governance and compliance in Foundry.
Manage quotas and gateways
Use Quota to review model quota and access allocation or increase-request capabilities. Quota access can require permissions beyond a single project. See Manage quotas for Foundry resources.
Use Gateways to access gateway-management capabilities. For guidance on choosing and configuring Azure AI Gateway or Azure API Management, see Configure Azure AI Gateway in your Foundry resources.
Quota management and gateway traffic controls are different mechanisms. Configuring a gateway isn't required to use every native Foundry quota, cost, or governance capability.
From fleet-wide insights to project-level investigation
Fleet-level and project-level views serve complementary purposes.
| Experience | Role |
|---|---|
| Manage | Broader inventory, operations, costs, governance, and administration. |
| Project-level monitoring | Detailed operational information and available traces for a project's agents and models. |
| Build | Development and configuration of AI assets within applicable organizational boundaries. |
| Integrated services | Specialized enterprise agent management, traffic governance, data protection, or threat investigation and response. |
For example, a platform team establishes approved tool and model requirements. Engineers configure agents within those boundaries. An operator notices increased latency in Fleet, investigates the affected agent in Operations, and follows project-level traces where available. If a separate data protection or security concern arises, the team investigates it through the configured Purview or Defender integration.
Compare resource identities, scopes, and time ranges across views. The pages aren't necessarily one automatically correlated incident, and trace availability depends on collected telemetry and permissions.
For setup and investigation guidance, see Set up tracing and configure fleet monitoring.
Get started
Begin with Foundry's built-in governance, then add integrations for the requirements that need them.
- Establish identity and access boundaries. Review agent identity and the identity and access overview. Grant only the management, endpoint, and downstream permissions your scenario requires.
- Configure supported governance and safety controls. Use model deployment policies, agent deployment policies, native tool governance, and guardrails as applicable. Verify configuration checks and runtime behavior separately.
- Establish cost visibility and controls. Review cost permissions, inspect subscription costs, and configure cost-control threshold actions for the usage you intend to cover.
- Inspect the resulting estate. Open Manage in the Foundry portal. Start with Fleet, then use Operations, Cost, or Compliance to investigate.
- Extend where needed. Add gateway controls, Agent 365 integration, Purview data protections, or Defender threat protection, following each service's coverage and configuration requirements.