unified-router

What problem does it solve?

The unified-router removes ad-hoc and inconsistent engine/model selection by automatically choosing the optimal LLM engine, model, and thinking depth for each task across Claude Code and Qwen Code, reducing cost, risk, and coordination errors.

Core Features & Use Cases

• Signal-driven routing: Chooses engine based on authority level, risk tier, MCP server needs, file coordination, and cost sensitivity.
• Model & depth selection: Selects appropriate models (e.g., opus/sonnet/haiku or qwen3.5/qwen3) and thinking budgets for architecture, implementation, or review tasks.
• Bidirectional dispatch: Supports offloading from Claude→Qwen, escalation from Qwen→Claude, and new dispatch flags like --think and auto activation for structured handoffs.
• Use Case: Automatically route a multi-file architecture plan to Claude opus while offloading single-file test generation to Qwen with --think to save tokens.

Quick Start

Ask unified-router to evaluate and dispatch the current task to the optimal engine and model with thinking enabled.