Relationship between ACP and AHP

I read the document AHP and the Agent Client Protocol
https://microsoft.github.io/agent-host-protocol/guide/ahp-and-acp.html
and I’m still confused about how AHP and ACP are intended to compose.

My current understanding is:

• ACP is designed for communication between a client and an agent runtime.
• Many existing clients (IDEs, editors, desktop apps, etc.) already implement ACP directly.
• AHP introduces a host that sits “between the client and the agent”, and the document says that clients send requests to the host via AHP, and the host translates them into ACP to talk to the agent.

This makes it sound like AHP replaces ACP as the protocol that clients should speak, which seems surprising, because it would require all existing ACP clients to adopt AHP as well.

An alternative interpretation would be:

UI Client  ←ACP→  AHP Host  ←ACP→  Agent Runtime

In this model:

• Clients continue to speak ACP (as they already do today).
• The AHP host acts as a virtual ACP client toward the agent runtime.
• AHP is used only between UI clients and the host, mainly for multi‑client state synchronization.

This seems more consistent with ACP’s existing ecosystem and with AHP’s goal of supporting multiple clients sharing a single agent session.

Could you clarify which of these interpretations is correct, and whether AHP is intended to replace ACP for clients, or to sit above ACP as a coordination layer?

Read the linked “AHP and the Agent Client Protocol” guide and compare its request flow with the two interpretations described in the issue. Confirm the intended relationship between the UI client, AHP host, and agent runtime with maintainers, then clarify the guide accordingly. The work is done when the client-facing protocol choice and composition model are unambiguous.