Agent Router

Route prompts to the optimal AI model based on task characteristics.

What This Does

Analyzes a task/prompt and recommends the best model, considering task type, complexity, cost, speed, and context length.

Model Routing Table

Instructions

1. Analyze the task: Classify by type, complexity (low/medium/high), and constraints (cost, speed, context length).
2. Check user preferences: Factor in model preferences, cost limits, or subscription tiers.
3. Route decision: Using the table above, recommend primary + fallback model.
4. For Codex CLI routing (local dev):
   • Heavy coding → codex exec (GPT-5.3-Codex, free via subscription)
   • Quick questions → current model
   • Cost-sensitive batch work → claude-haiku or gpt-4.1-mini
5. Output format:

   🔀 Route: <task_type> | Complexity: <level>
   ✅ Recommended: <model>
   🔄 Fallback: <model>
   💰 Est. cost: <low/medium/high>
   💡 Reason: <why>
   

6. Batch routing: For multiple tasks, create a routing plan table showing which model handles each task.

Edge Cases

• Ambiguous tasks: Default to claude-sonnet as the best general-purpose choice
• Multi-modal tasks (text + image): Ensure the chosen model supports vision
• Context overflow: If input exceeds model's context window, suggest chunking or switching to a larger-context model
• Rate limits: If primary model is rate-limited, auto-route to fallback

Requirements

• No API keys or dependencies — this is a decision framework, not an API
• Update routing table as model pricing/capabilities change