jargon

Applied AI·Production patterns and cost

you route every model call through one service so keys, logging, caching and cost attribution live in one place instead of scattered through the codebase.

LLM gateway / proxy

Also calledproxy

Draft summary, pending review

A single service through which all model calls flow, providing keys, logging, caching, rate limiting, routing and cost attribution in one place. Exactly an API gateway with token-aware features; buy one or write a thin one, but do not let calls scatter unmediated through the codebase.