Not one model. Not one vendor. One owned runtime.

This session shows how to build an owned agent runtime using Microsoft infrastructure, OpenClaw, and Waza. The reference use case is Tula, an open-source personal health AI built as OpenClaw skills.

We will walk through the architecture for an enterprise-controlled agent running on Azure, using Entra ID and Microsoft Graph for Microsoft 365 integration, OpenClaw for the agent runtime, Tula skills for inspectable domain behavior, and Waza evals for behavioral verification.

The key lesson is not “use one model” or “use one vendor.” The opposite is true, real agents are heterogeneous. Enterprise teams need to own the runtime, scope the permissions, isolate the workspace, and verify the behavior before they trust the agent inside business workflows.

Attendees will leave with a reference pattern for Microsoft-centered agent systems that are inspectable, self-hosted, and testable.