Model Routing and Failover Runbook

Model routing is an operational decision. A cheap model may be perfect for a summary, while a high-stakes code review, brand asset, or public deployment decision may deserve a premium model.

Official Hermes Agent docs: https://hermes-agent.nousresearch.com/docs/integrations/providers

Routing checklist

• Match model quality to task risk and creative requirements.
• Prefer provider paths that are already configured and verified.
• Use premium models for client-facing copy, security review, complex coding, and visual quality calls.
• Use cheaper or default tiers for routine summaries, background scans, and low-risk data extraction.
• When a provider fails, retry or choose an approved fallback rather than hiding the failure.

Failover standards

• Record the provider failure in the final report when it affected scope.
• Do not substitute a different image provider when the user required Banana or Gemini unless they approved it.
• For code work, rerun verification after switching models or strategies.
• For public facts, use web or docs checks regardless of model confidence.
• Keep generated claims grounded in tool output or official sources.

Pitfalls

• Burning premium capacity on simple mechanical checks.
• Using a low-tier model for a high-stakes public launch review.
• Claiming an image or deploy succeeded when the provider call failed.
• Confusing model availability with tool availability.

Verification steps

• Confirm configured provider status when model routing is the task.
• Check output quality against acceptance criteria, not just completion.
• Run real tests or live probes for any model-produced code or content.
• State remaining provider limitations clearly.