Manifest Cost Reduction Flow

Cheaper default routing, premium escape hatch, cache stability, and zero-cost local fallback

PRIMARY SAVINGS = CHEAP DEFAULT ROUTE + STABLE CACHE Switch UI same chat workflow same provider picker cost change happens underneath Hermes Agent reads config.yaml + .env Provider split manifest-cost = default work manifest-premium = expensive tasks only auxiliary tasks can route separately manifest-cost cheap default route normal chat, coding, daily tool use main savings path manifest-premium premium route vision, compression, deep synthesis use sparingly Cheaper backend lower-cost OpenAI-compatible endpoint hosted, self-hosted, or internal gateway Premium backend frontier quality for selected tasks higher cost, higher value Prompt cache hits keep model, tools, and system prefix stable avoid mid-session provider churn cheaper repeated turns Delegation + compression delegate_task → smaller isolated contexts /compress → shrink long-session history execute_code → batch operations less token duplication Local fallback Ollama or local OpenAI-compatible server best for routine low-value work zero per-token cloud cost default route premium-only tasks stable sessions reduce context cost

Cost controls

  • • Cheap default route via manifest-cost
  • • Premium route only for selected auxiliary tasks
  • • Same UI, different billing path underneath
  • • Centralize routing behind one endpoint

Token savings

  • • Stable provider/model keeps prompt cache warm
  • • /compress reduces future turn cost
  • • delegate_task keeps subwork in smaller contexts
  • • execute_code avoids many tiny tool loops

Zero-cost path

  • • Local Ollama for routine work
  • • No per-token cloud billing
  • • Best when quality needs are moderate
  • • Good complement to Manifest routing

Hermes Switch UI • How-to diagram for Manifest provider cost reduction