Model Routing System

What problem does it solve?

It solves the problem of getting consistently good, cost-aware answers by automatically classifying each user prompt and delegating it to an appropriate model with cascading fallbacks and verification when the classifier or tools fail.

Core Features & Use Cases

• 9-tier cost/complexity routing: assigns a tiered model choice based on prompt type, including cheap local classification first and escalation to higher-quality models when needed.
• Classifier fallback + cached recovery: retries classification across multiple backends and uses a routing cache so routing continues even if Ollama is temporarily unavailable after context compaction.
• Tool fallback protocol with research evaluation: swaps to external backends (e.g., multi-backend research) and can auto-evaluate results to select better-performing options.
• Integration with Maggy routing rules: enforces task-type overrides (e.g., security/architecture to Claude, bulk to Gemini Flash-Lite) while mirroring the same tier structure in Maggy.

Quick Start

Ask Claude Code to handle your request normally and let the route-task hook classify the prompt, delegate to the selected tiered model, and fall back automatically if routing or tools fail.