01 / MODEL
Model routing
Compare only approved models by quality, context, latency, cost, and tool-use needs. Escalate when no candidate is safe enough.

Task, real candidates, permissions, risk, and current evidence.
Choice, Score, or Noul returns typed answers and probabilities.
Allow, confirm, review, or stop without expanding authority.
Your Agent host executes and records the append-only receipt.
Compare only approved models by quality, context, latency, cost, and tool-use needs. Escalate when no candidate is safe enough.
Put a typed decision in front of Skills, MCP tools, and CLIs. Preserve permissions, risk, confirmation, and execution ownership.
Choose the specialist for the current state, then hand execution back to your Agent host with a receipt you can inspect.