microsoft-foundry

Agent lifecycle management in Foundry covers building, deploying, evaluating, and monitoring both hosted and prompt-based agents. It coordinates pipelines with RBAC permissions and dataset evaluator workflows across dev and production environments.

RBAC and resource provisioning are integrated with Foundry projects to manage permissions and allocate resources across environments. This ensures agents have appropriate access controls during deployment and evaluation workflows.

Batch evaluations can be run on deployed agents using dataset and evaluator workflows within Foundry. This allows monitoring metrics across environments after onboarding agents to a dev project and deploying them through standard pipelines.

Foundry deployment supports both hosted and prompt-based agents through a unified lifecycle framework. The standard agent setup workflow handles deployment, invocation, and observation across agent types via sub-skills like deploy, invoke, and observe.

Onboarding a new agent to a Foundry dev project uses the standard agent setup workflow to bootstrap the project and deploy the first hosted agent. This coordinates RBAC provisioning, metadata validation, and integration with Foundry MCP for end-to-end governance.