Multi-Comfy Orchestration

Run several ComfyUI instances on mixed local hardware and pick the right one per job. Typical shop: NVIDIA DGX Spark (planning / Flux / Z-Image) plus dual RTX 3090s (one GPU per Comfy process, image vs video).

An isolated Hermes agent (isolated-hermes-home) is the router. It must not assume a single localhost:8188.

When to Use

• More than one ComfyUI HTTP port is live (8188, 8189, …)
• Hardware mix: DGX Spark + 3090s, or two 3090s, or Spark-only
• Queueing stills and LTX/Wan video without stalling one box
• User mentions Spark, 3090, multi-instance Comfy, or "fully local"

Don't use for: a single hobby Comfy on one port (use comfyui instead).

Prerequisites