Modelocale vs one-shot MT

Raw machine translation on every request is simple — and expensive, slow, and inconsistent.

One-shot MT

Typical loop:

request → translate whole payload → respond

Problems:

• Latency and MT cost on every hit
• No memory — wording can drift run to run
• No shared key with your routes — agents invent resource ids
• Hard to skip SKUs, ids, and brand strings cleanly across calls

Modelocale

ensure(endpoint_url, lang, data)  → remember localized JSON
get(endpoint_url, lang)           → cache-fast read, no MT

You still use MT — but only on misses and source changes, not on every GET. The identity is your URL (endpoint_url), scoped to your project.

That matches how vibe coders and agents already think: “give me /v1/menus in fr.”

When one-shot still wins

Throwaway scripts, one-off docs, or content you will never re-fetch. For a live API that serves the same locales repeatedly, ensure/get is the better fit.

More detail: Concepts.