Priced on what actually costs us money.

Because a document is unbounded. A one-kilobyte note and a 400-page scanned manual are both "one document", but the manual costs roughly 400 times as much to parse, OCR, embed and store. Per-document pricing lets one customer subsidise another by orders of magnitude and invites gaming by concatenation. A page is the parser's real page count where the format has one, otherwise about 3,000 characters of extracted text — measured on extracted text, so you never pay for markup or whitespace.

Embedding tokens and reranking calls, but only when troveGEN supplies the model. On BYO Vector DB you connect your own OpenAI or Cohere key and those calls are metered at zero here — you pay your provider directly. troveGEN-Z is fully managed, so every model is ours and these rates always apply. Published rates are $0.30–$0.60 per million embedding tokens, $4.00–$7.00 per thousand reranks, and $3.00–$6.00 per million generation tokens. Reranking is priced per search query rather than per token, and is the single largest model cost we carry — the published floor sits above what a cross-encoder provider charges us, so no tier is sold below cost.

It comes down to one question: do you want to own the vector database? If yes, take BYO Vector DB — you run the store, connect your own embedder, reranker and LLM if you want them, and pay nothing here for models you supply. If you would rather not operate any of that, take troveGEN-Z, where we host the vectors, the embedding model, the reranker and the LLM. A workspace holds one subscription, so the lines never stack, and the free trial grants both paths so you can try the one you intend to buy.

Yes, because it is the most expensive step in the pipeline by an order of magnitude. Each plan includes an OCR page allowance, with overage at $6–$10 per 1,000 pages. You are billed for pages actually recognised, not the document's page count — a capped run reads fewer, and charging for pages we never looked at would be wrong. You can also set OCR to off, in which case a scanned file is rejected with a clear reason rather than silently indexed as an empty shell.

The request is refused with a clear message naming the metric and the number. There is no silent metered overage — we do not have mid-cycle invoicing, and quietly running up a bill we cannot invoice would be worse than an honest stop.

Yes. One global list price in USD, wherever you are. Data residency is a deployment choice on the Sovereign tier, not a pricing tier.

Yes, and upgrading from the free trial carries your existing corpus across into your new dedicated schema in a single transaction — your documents are never in two places or neither. Annual billing is ten times the monthly price, so two months are free.