Pricing
Priced by the work.Quoted after the inventory.
Every data set is inventoried before it is quoted. The inventory takes minutes, costs nothing, and tells both of us the size, the formats and the surprises. Unusual collections are quoted after that step, never before.
Paid once per data set. Then, if you want it kept current, monthly.
- ✓Every document extracted, cleaned, chunked and deduplicated
- ✓Processing report and flag lists
- ✓Curator's review and sampling
- ✓Thirty-minute decision session
- ✓Signed decision record
- ✓Ten test questions, answers and sources logged
- ✓Your assistant, search or AI platform
- ✓Retrieval that cites sources
- ✓Instructed to say “not listed” rather than guess
- ✓Delivery test against the connected assistant
- ✓Scheduled refreshes, monthly or on demand
- ✓Change report: new, changed, removed chunks
- ✓Re-indexing and re-run test questions
- ✓Answer-quality monitoring
- ✓Quarterly review
Add-ons, only when the collection needs them: OCR for scanned archives, additional languages, on-premise processing, compliance documentation for the AI Act.
Tell us about your collection.
Choose a package, answer five questions, and you hear from us within two working days. Nothing is uploaded here; the inventory comes after we talk.
Request received.
A confirmation is on its way to your inbox. If you already have the folder or a file list, reply to it and we start the inventory straight away.
The inventory places a data set in a band.
Band A
Text PDFs and Word documents with a text layer. No spreadsheets, no scans. The fastest turnaround.
Band B
Spreadsheets and directories, legacy Office formats, more than 500 files, or person-level columns that need a personal-data decision.
Band C
Scanned pages that need OCR, multi-column magazine layouts, protected files, more than 2,000 files or more than 2 GB. Quoted individually.
Band prices are being set with the first pilots, on the curation hours they show. Every quote states its band, what is included, and what “complete” means.
A delivery is complete when all of this is true.
- Every line of the knowledge base is valid, every chunk carries its source, page or sheet, and date; no empty chunks, no duplicates, spreadsheet values still tied to their column names.
- Every unreadable file is listed with its reason. Every duplicate and every kept version is recorded.
- Every personal-data, version, stale-document and conflict decision is written down with your sign-off person's name and the date of the session.
- Ten test questions whose answers exist only in your documents have been asked to the connected assistant; the answers and their sources are attached.
- A refresh reproduces the same output from the same documents, and reports exactly what changed when they change.