Workers AI as a queued tier over the queue the HTTP and TCP tiers use.
Queued rather than parked: a generation takes seconds and the neuron meter is a hard 429, so an
inline call would fail a page for a quota it did not need. Streaming stays closed (a park
delivers one answer). Uses the AI binding, never REST, so no account token sits in a queue
row. Neurons (10,000/day, reset 00:00 UTC) are projected; with no binding every call refuses.
Workers AI as a queued tier over the queue the HTTP and TCP tiers use.
Queued rather than parked: a generation takes seconds and the neuron meter is a hard 429, so an inline call would fail a page for a quota it did not need. Streaming stays closed (a park delivers one answer). Uses the
AIbinding, neverREST, so no account token sits in a queue row. Neurons (10,000/day, reset 00:00 UTC) are projected; with no binding every call refuses.