Infrastructure

Most corpora share a retrieval pool the platform runs for you and never need anything here. This is for the ones that outgrow it: a corpus that is large, latency-sensitive, or wants isolation can be placed on a dedicated retrieval resource you lease by the hour. list shows the catalog of sizes and the resources you already hold, get opens one, and logs streams what its instance is doing. provision leases a size, resize moves a resource to another size, and rename labels it.

A resource you are not using can hibernate: it drops to a holding price and keeps nothing resident, since its sources' documents live in the platform's own store either way. resume brings it back. destroy returns it, though it is refused while any data source is still placed on it. Each size is quoted by how many chunks it holds at a given embedding dimension, so a resource's capacity depends on what is placed on it.