Model Router Skill

Intelligent task routing that spawns subagents with optimal models based on task complexity.

Philosophy

• Main session: Always fast (llama-3.2-1b) for instant responses
• Subagents: Spawned with task-appropriate models
• Parallel execution: Multiple complex tasks can run simultaneously

Model Tiers

Tier	Model	Alias	Best For
Fast	nvidia/meta/llama-3.2-1b-instruct	fast	Chat, simple QA, summaries
Balanced	nvidia/minimaxai/minimax-m2.1	balanced	General tasks, analysis
Advanced	nvidia/moonshotai/kimi-k2.5	advanced	Complex reasoning, long context
Reasoning	nvidia/deepseek-ai/deepseek-r1-distill-qwen-32b	reasoning	Math, logic, deep analysis
Code	nvidia/qwen/qwen2.5-coder-32b-instruct	code	Programming, debugging

Commands

Route Task

route [task] with [model]
route analyze XAUUSD strategy with advanced
route debug this Python error with code
route deep research on quantum computing with reasoning

Auto-Route (Smart Detection)

auto-route [task]

Automatically detects complexity and spawns appropriate subagent.

Spawn with Model

spawn [task] --model [alias] [--timeout N]

Task Complexity Detection

The router auto-detects task type:

Fast (no spawn):

• Simple greetings
• Quick questions
• Status checks
• Acknowledgments

Balanced:

• Market summaries
• Strategy explanations
• General analysis

Advanced:

• Multi-step analysis
• Trading strategy development
• Complex calculations

Reasoning:

• Mathematical proofs
• Logical puzzles
• Deep analysis requiring step-by-step thinking

Code:

• Programming tasks
• Debugging
• Code review
• Technical implementation

Usage Examples

Trading Analysis

You: Analyze XAUUSD breakout strategy
Router: [spawns subagent with advanced model]
Subagent: [performs deep technical analysis]
Router: [presents summarized results]

Coding Task

You: Fix this Python script for data processing
Router: [spawns subagent with code model]
Subagent: [debugs and refactors code]
Router: [presents solution]

Simple Chat (No Spawn)

You: What's the weather?
Router: [handles directly with fast model]

Workflow

1. User sends request
2. Router analyzes task complexity
3. If complex → spawn subagent with optimal model
4. Subagent performs work in background
5. Router presents results when complete

Configuration

Override model mapping in config:

{
  "modelRouter": {
    "tiers": {
      "fast": "nvidia/meta/llama-3.2-1b-instruct",
      "balanced": "nvidia/minimaxai/minimax-m2.1",
      "advanced": "nvidia/moonshotai/kimi-k2.5",
      "reasoning": "nvidia/deepseek-ai/deepseek-r1-distill-qwen-32b",
      "code": "nvidia/qwen/qwen2.5-coder-32b-instruct"
    },
    "defaultTimeout": 120,
    "maxConcurrent": 4
  }
}