zedarvates Tiered Model Router (P14)
5-level intelligent model selection with cost estimation and auto-downgrade.
Trigger: Before EVERY LLM call — route through tiered router first.
5 Levels:
L0 FREE (0 tok): Hailo-8, LocalAI TTS/STT, pure math/vectors
L1 LOCAL (~100 tok): LocalAI Gemma-4 / Ollama — simple Q&A
L2 CHEAP (~500 tok): Cloud small — code review, bug detection
L3 STANDARD (~2K tok): Cloud standard — architecture, complex reasoning
L4 PREMIUM (~8K tok): Cloud best — security audit, system design
Features:
- Cost estimation before call (tokens + $)
- Auto-downgrade when budget exceeded
- Per-project daily/monthly budget limits
- Usage tracking + savings report
- Agent-to-agent delta compression
Module: skills/tiered_router
Typical savings: 95-99% vs all-PREMIUM