Use case · Document search
Find the right information in your documents, with its sources.
A document assistant (a technique known as RAG) searches only your authorised documents, cites the passages it finds and builds its answer from them. It does not answer from memory and does not guess.
- Your documents
- Question
- Answer with sources
- Sources
- Procedures, standard contracts, product sheets, technical documentation.
- Answers
- Always accompanied by the cited passages.
- Deployment
- Cloud, private cloud or fully local.
A general-purpose assistant answers with what it learned during training and may confidently invent a detail. A document assistant searches a defined set of documents, finds the relevant passages and answers only from what it has found.
If nothing relevant exists in the authorised sources, the right answer is “I cannot find this information”, not an invented answer.
How it works
Four verifiable steps.
- 01
Question
An employee asks a question in everyday language, for example: what is our refund procedure for a business customer?
- 02
Search in authorised sources
The system searches only the documents you have explicitly included, not the internet or other folders.
- 03
Answer with citations
The answer is built from the passages found, with a reference to the exact document and location.
- 04
Human approval
For sensitive use, such as a reply to a customer or a contractual question, a person reviews the answer and its sources before it is shared.
Permissions and sources
The assistant sees only what it is allowed to see.
The document scope and access rights are defined before go-live, never left to defaults.
| Item | What is defined | Why |
|---|---|---|
| Included sources | An explicit list of folders, wikis or document repositories | To prevent a sensitive or outdated document from being indexed by mistake |
| Access rights | A user retrieves only the passages they are authorised to view | To respect existing internal separations: HR, finance, clients |
| Updates | Scheduled re-indexing, or re-indexing triggered when a source changes | To avoid answering with an outdated version of a procedure |
| Log | History of questions asked and sources cited | To allow later review |
Frequently asked questions
Document assistant, without shortcuts.
Can the assistant make up an answer?
The risk exists with any system based on a language model. It is greatly reduced by systematically citing sources: the answer can be checked in one click, and the absence of a relevant passage must result in “I cannot find it”.
Do we need to index all our documents from the start?
No. We recommend starting with a limited, stable scope, such as one procedure or a product database, checking the quality of the answers, then expanding gradually.
Are access rights respected?
Yes. Existing permissions, by team or by role, are applied at search level, not just at interface level.
Is it compliant with the Swiss FADP or the GDPR?
No architecture is compliant by default: compliance depends on the data processed, the hosting and the use. The Federal Data Protection and Information Commissioner (FDPIC) points out that the FADP applies directly to AI-based data processing. We scope these points with you.
Can everything be kept local?
Yes, with a private AI architecture: model, index and documents remain on infrastructure you control.
Document search
Information your team spends too much time looking for?
Tell us where it is today. We will propose a first document scope that can be tested.