1 · Traceable
Answers are designed to cite the document passages they used — for your verification.
Ask your documents. Every answer comes with its source.
We turn your document corpus into a private search-and-answer base: answers link to the passages they came from, and return "not found" when your documents don't support them.
your documents · traceable answers · founding-client rates
THE DIFFERENCE
Answers you can trace
2 · Coverage checked
Citation coverage is verified programmatically. It helps reduce hallucinations; it does not guarantee accuracy.
3 · Honest "not found"
When the system cannot find support in the indexed sources, it returns a not-found answer instead of an unsupported one.
SAMPLE CORPUS PREVIEW
See it answer from a sample corpus
Try a sample question below against a small fictional document set. This preview shows how the system finds and cites an answer — not a live connection to your documents.
Response will appear here once you send a request.
SAMPLE ANSWERS
What a traceable answer looks like
The useful part is not that the system answers quickly. It is that the answer shows where it came from, and refuses to answer when the documents do not support it.
CITED ANSWER
Find the termination clause in our contracts
Knowledge Studio · risk: LOW
Found in 2 of 6 indexed agreements (sample corpus):
• 60-day written notice, either party. [msa-2025.md, §11.2]
• 30-day notice during the pilot period. [pilot-addendum.md, §4]
The other four agreements contain no termination clause in the indexed text.
HONEST NOT FOUND
What's our policy on remote work?
Knowledge Studio · risk: LOW
Not found in the indexed sources (sample corpus).
Nothing in these documents covers a remote-work policy. The system returns "not found" rather than assembling an answer the sources don't support — that is the intended behaviour, not a failure.
Acceptance checks before delivery
- Answers must cite indexed source documents, not general memory.
- Questions outside the corpus must return a not-found answer.
- The client can open the cited source and verify the passage before relying on it.
Illustrative sample corpus, not a fake client archive. A real knowledge base is tested on your documents and your real questions before acceptance.
WHO THIS IS FOR
Who this is for
Teams sitting on documents nobody can search — contracts, procedures, technical archives, years of internal know-how.
Knowledge nobody can find
Answers exist somewhere in the archive; finding them takes an afternoon.
Answers that need a source
In regulated or audited work, an answer without a reference is worthless.
Onboarding takes forever
New people re-ask the same questions the documents already answer.
Corpora too sensitive for a public cloud
Documents you would not paste into a public AI tool.
Typical corpora
Contracts & agreementsInternal proceduresTechnical documentationAccounting archivesProject histories
Working with legally protected secrets (secret professionnel)? Those need network isolation — see the Private AI Workstation.
WHAT YOU GET
What you get
- Corpus review & written scope
- Your documents indexed — text directly, scanned files via OCR where quality allows
- A private search-and-answer interface
- Answers with links to the passages used
- Handover + training
- Optional care plan — remote-first support, by appointment
How we work
- 01Scoping call (20 min)
- 02Corpus review + fixed quote
- 03Index & build
- 04Acceptance test + training
- 05Care plan (optional)
WHERE IT RUNS
Where it runs
On your machine
Local-first by default: the corpus and the index stay on infrastructure you control.
On an existing setup
If you already have suitable infrastructure, we deploy there. Scope confirmed on the call.
"Infrastructure you control" is not the same as network isolation. Any external service, update channel or support access is documented before work starts. If you need a fully isolated, air-gapped setup, that is the Private AI Workstation.
PRICE & ACCEPTANCE TEST
Book a scoping call
Small corpus pilot from 1 900€ · most complete deployments 2 900-4 500€ after corpus review
OCR, unusual formats, integrations or extra access controls can change the scope — the quote is fixed once the corpus is reviewed.
Founding-client rates — reduced while we build the public portfolio. Limited slots.
The acceptance test
Before we build, we agree an acceptance test in writing, with a verifiable criterion — for example: on a set of your real questions, the answers cite the correct source document. If it doesn't meet the agreed criterion, you don't pay the implementation fee. Accuracy of every future answer is not guaranteed — that's why answers are traceable.
FAQ
Common questions, honest answers
Do my documents leave my machine?
The default design is local-first: the corpus and index stay on infrastructure you control. Any external service or support access is documented before work starts.
What if the answer isn't in my documents?
It returns "not found". The system is built to report the absence of support rather than produce an unsupported answer, and citation coverage is checked programmatically before acceptance.
Can I delete specific documents from the index later?
Yes. You remain the data controller: we implement deletion of specific documents or records from the index on your request, and the procedure is written into the DPA we sign before any work on your data.
Does it handle scanned documents?
Text documents index directly; scanned files go through OCR where the quality allows. We confirm what's feasible on your actual corpus during the review — before the quote is fixed.
Does this make me GDPR-compliant?
No tool makes an organisation GDPR-compliant by itself. It can support data minimisation and local processing; you remain the data controller and must have a lawful basis for the personal data in your documents. Castel Studio acts as a data processor under a DPA signed before any work.
Who can see the answers?
Access is configured for the users you approve on your own system; the final access control depends on the agreed deployment. We don't host your corpus.
What this is not
- Not a guarantee of accurate answers — answers are traceable so you can verify them.
- Not an air-gapped solution — if you need complete network isolation for legally protected secrets, that's the Private AI Workstation.
- Not a search engine over the internet — only the documents you provide.
- Not a compliance service — GDPR obligations stay with you.
Book a scoping call
Book a scoping call
Prefer email? contact@castel.studio
Indexing may involve personal data. You are the data controller; Castel Studio acts as processor under a DPA signed before any work on your documents. You must have a lawful basis for processing the personal data they contain. Deletion from the index is supported on request.