Fleet Status

Local model fleet snapshot — per-machine reachability, currently loaded model(s) with live context utilization, and hardware (GPU/VRAM/RAM), across the local-inference fleet (LM Studio + LM Link).

MCP name: fleet_status

Category: inference

Details

Returns the local model fleet snapshot: one entry per machine in the local-inference fleet, with reachability, the model(s) currently loaded (with quantization, status, and live context length vs the model's trained max), and hardware (GPU / VRAM / RAM). Answers "what's running on which box" — the per-machine question the inference broker can't, since the broker only knows the profile (model), not the machine.

Takes no parameters. Reads are live (the configured local-inference provider — LM Studio today — is enumerated via its CLI) and never raise: when the provider is unreachable, each machine degrades to reachable: false and lms_available is false. Remote-peer hardware comes from the inference.fleet config roster (the local machine reports its own hardware live); each machine's hardware.source records whether it was read live or came from the roster.

Use this to decide where a model should run — e.g. whether a model that won't fit the local GPU should be routed to a higher-VRAM peer. The same snapshot backs the dashboard's Settings › Inference fleet section.