Why This Matters

Different file types benefit from different models. A router optimizes cost and quality simultaneously.

Concept Explained

Router: input file type → pick model. TypeScript → GPT-5 (strong type awareness). Python → Claude (good idioms). Markdown → GPT-5 mini (cheap). Etc.

How It Works

Build router that takes file type (or detects from extension), picks model based on config, calls Copilot. Track cost savings.

Step-by-Step Tutorial

1. Define routing rules

2. Build router

3. Add fallback

4. Track cost

5. Adjust based on data.

Real-World Example

A team built a router: TS→GPT-5, Py→Claude, MD→mini. Cost dropped 35% (cheap model for docs), quality maintained (right model per type). Adopted org-wide.

Example Prompts / Commands / Code

"""# router-config.json
{
  "routing_rules": {
    ".ts":  "gpt-5",
    ".tsx": "gpt-5",
    ".py":  "claude",
    ".go":  "claude",
    ".md":  "gpt-5-mini",
    ".json": "gpt-5-mini",
    ".yml": "gpt-5-mini",
    "default": "gpt-5"
  },
  "fallback_chain": {
    "gpt-5": ["claude", "gpt-5-mini"],
    "claude": ["gpt-5", "gpt-5-mini"],
    "gpt-5-mini": ["gpt-5", "claude"]
  }
}
"""

"""import os
import json

class ModelRouter:
    def __init__(self, config_path='router-config.json'):
        with open(config_path) as f:
            self.config = json.load(f)

    def pick_model(self, file_path: str) -> str:
        ext = os.path.splitext(file_path)[1]
        return self.config['routing_rules'].get(ext, self.config['routing_rules']['default'])

    def route_with_fallback(self, file_path: str, prompt: str) -> str:
        model = self.pick_model(file_path)
        for m in [model] + self.config['fallback_chain'].get(model, []):
            try:
                return copilot.run(prompt, model=m)
            except:
                continue
        raise RuntimeError('All models failed')

# Usage:
router = ModelRouter()
result = router.route_with_fallback('src/auth.ts', 'refactor this function')
"""

Common Mistakes

• Routing too fine-grained — complexity without benefit.
• No fallback — single model failure breaks system.
• Not tracking cost savings — can't justify router.
• Not adjusting based on data — routing stays static.

Best Practices

• Route by file type (5–10 categories).
• Build fallback chain.
• Track cost savings vs single-model baseline.
• Adjust routing based on data.
• Start simple; add complexity as justified.

Troubleshooting

Practical Exercise

Professional Challenge

Frequently Asked Questions

Further Reading

• Model comparison

Official References

• Model comparison

Related lessons: PR-05, PR-50