hermes - 💡(How to fix) Fix Implement Resource-Aware Model Fallback Chain (Local → Remote via LM Link)

Problem: When the same model is available both locally and via LM Link, Hermes always prefers the local endpoint. If local resources become constrained (high CPU/memory), there's no mechanism to automatically fall back to the remote LM Link copy.

Desired behavior:

• Monitor local resource utilization (CPU, memory, GPU) during inference
• Implement resource-aware fallback: if local resources exceed a threshold, automatically route requests to the same model via LM Link
• Support configuration of resource thresholds per model or globally

Example config:

model_fallback:
  - local_llm:1234
  - lm-link://remote-ip:1234
resource_threshold:
  cpu: 80%
  memory: 85%