warm

Seeds the cache with known prompt/response pairs, embedding them in one batch call where the Embedder supports it (see Embedder.embedAll).

The intended use is startup warming from a fixed set — an FAQ, a golden set of canned answers — so the first real user of each prompt gets a hit instead of paying to populate the cache. It is far cheaper than putting them one by one: a provider that prices and rate-limits per request embeds the whole batch for the cost of a single round trip.

Entries are written in the given order and each gets a fresh id, exactly as put would; the returned ids line up with entries. Warming does not consult or move any counter — it is a write path, not a lookup — and it bypasses the negative cache. An empty entries is a no-op.