Route easy prompts to cheap models, automatically

"Summarize this ticket" does not need the same model as "find the race condition in this diff." But your app has one configured model, so both pay frontier prices.

A hoody-exec MITM fixes that at the traffic layer. The script inspects each request: length, task shape, a cheap heuristic or a fast classifier: and rewrites the model field before forwarding to the Hoody AI gateway: small model for routine calls, frontier model for the hard ones. The app still thinks there's one model. The gateway bills your AI Balance for what each call actually used.

Routing this way cuts token spend 40-70% on mixed workloads, and no application code changed.