An agent that knows what it does not know.

Design principles

• Cite or decline. If a claim cannot be cited to a Framework section or PMO Delivery Desk page, the agent does not make the claim.
• Retrieval over generation. The agent's job is to get the practitioner to the right page, not to replace it.
• Authority remains in the corpus. The agent is a delivery surface, not a policy author. It never invents standards.
• Bounded scope. Out-of-scope questions are routed to the right human or politely refused.

What it does

Practitioners ask the agent questions in natural language. The agent retrieves the relevant PMO Delivery Desk page or Framework section, summarizes the answer, and provides a clickable citation. For procedural questions, the agent walks the user through the JTBD steps from the corresponding Delivery Desk page.

For policy questions, the agent cites the Framework section directly. Practitioners can move from procedure to policy and back with consistent grounding.

Built on Copilot Studio

The agent is delivered as a Copilot Studio configuration: topics, prompts, knowledge sources, and orchestration logic. Customers deploy it inside their Microsoft 365 tenant. Microsoft Teams is the primary surface. SharePoint embedding is supported with appropriate DLP exception handling.

This is intentional. Customers run the agent under their own identity, security, and compliance posture. Clearline supplies the pattern. The customer owns the deployment.

What it does not do

• It does not author policy
• It does not approve work
• It does not replace human judgment on tier classification, escalation, or risk acceptance
• It does not invent answers when the corpus is silent

This restraint is the feature. Practitioners trust the agent because it does not pretend to know things it does not know.