IICP now supports MeshLLM as a distributed inference backend

[RobLe3]

Hi MeshLLM maintainers and community,

We implemented MeshLLM support on the IICP side and wanted to validate that we
are using the intended public API boundary.

IICP is an open capability-discovery and routing protocol. An IICP provider can
use MeshLLM as its local inference runtime. The boundary is deliberately small:

IICP provider → MeshLLM OpenAI-compatible gateway → MeshLLM routing and execution

What the IICP integration does:

• checks MeshLLM readiness and the /v1/models inventory;
• forwards standard chat requests to a local http://127.0.0.1:9337/v1 gateway;
• advertises explicitly selected, stable text-chat models for IICP discovery;
• keeps the provider's operating region separate from MeshLLM internals; and
• excludes MeshLLM's experimental model: "mesh" ensemble alias from ordinary
  automatic capability advertisement.

What it deliberately does not do:

• modify MeshLLM core or join MeshLLM's control plane;
• expose MeshLLM peer, stage, route, or topology metadata through IICP; or
• treat the experimental ensemble alias as a stable physical model.

The initial IICP compatibility profile has been exercised against a real
MeshLLM gateway, and a MeshLLM-backed provider is currently visible in the IICP
directory. For a user-facing overview of the connection and how to try it,
see the IICP community post.

Could you let us know whether /readyz, /v1/models, and OpenAI-compatible
chat completions are the intended stable external boundary for integrations
such as this? We welcome corrections to any assumption we have made.