跳到主要內容

Integrate external agents from registry sync with A365 using A365 SDK

Implementation Effort: Medium – Requires developers to add the Agent 365 SDK to each external agent's codebase, configure Entra-backed identity and blueprint association, and validate governed data access, in coordination with the teams that own those agents.
User Impact: Low – Primarily an admin and developer activity; once integrated, agents gain identity and governance controls without changing how end users interact with them.

Overview​

Registry sync brings agents built on external platforms such as Amazon Bedrock and Google Vertex AI into the Agent 365 registry, but synchronized agents start with only a thin layer of governance — they are visible in the inventory and support a limited set of management actions such as deletion. They still authenticate with credentials managed outside Microsoft Entra, access organizational data through their own integrations, and operate without the observability and policy enforcement that govern native Microsoft agents. Visibility alone does not close that gap. Integrating these external agents with the Agent 365 SDK is the step that turns a discovered agent into a governed one. The SDK does not rebuild or rehost the agent; it extends an agent that already exists — regardless of the framework it was built on — with an Entra-backed agent identity, governed access to Microsoft 365 data through admin-controlled Model Context Protocol servers, full observability through OpenTelemetry, and the ability to operate inside an IT-approved blueprint that defines its compliance, security, and lifecycle policies.

This supports Verify explicitly by giving each external agent a first-class Entra identity, so that every action it takes is authenticated and authorized against the same identity, Conditional Access, and risk signals that apply to other agents, rather than trusting credentials issued and held outside the organization's control. It supports Use least privilege access because an SDK-integrated agent reaches Microsoft 365 data only through governed connections that an administrator scopes, instead of broad API integrations the security team cannot see or constrain. It supports Assume breach by making the agent's behavior observable and auditable end to end — traceable inference events and tool usage mean a compromised or misbehaving external agent can be detected and contained rather than operating silently. Without this integration, externally built agents remain partly ungoverned even after they appear in the registry: they hold standing access through credentials Entra does not manage, their data access cannot be narrowed to least privilege, and their activity produces no audit trail, leaving a blind spot that a threat actor can exploit without detection.

Reference​