Swapping models is editing a string

A new open model drops and you want it under your feature by Friday. On per-model inference platforms, each model is its own endpoint with its own shape: evaluating the field means integrating the field.

On Hoody, your container talks to one OpenAI-compatible gateway that fronts 300+ models across the open-model providers: Llama, DeepSeek, Qwen, Mistral. Moving your feature from one to another is editing the model string in the request. The auth doesn't change either: the container's own identity token, minted automatically, is the credential.

Your Friday eval harness calls the same URL five times with five model strings and diffs the outputs.

Trying five models is a for-loop, not five integrations.