Chat completions
The drop-in. Routes, caches, maybe escalates.
POST/v1/chat/completions
messages*array
OpenAI messages.
modelstring
auto (default) or a fleet id.
level_overrideinteger
Force complexity 1–5.
streamboolean
Must be false.
Extra response
usage.cost holds actual, baseline, saved, cache discount. promptimizer holds classification, tier, cache flags, escalated, quality_gate.