Security & deployment
Your documents, under your rules
Reperix is built for companies that cannot afford their documents ending up where they shouldn’t. Data and permission protection are at the core of the product, not an accessory.
Principles
Reperix’s security principles
-
Your data stays under your control
On-premise, your document corpus stays in your infrastructure. Analysis and search happen in your own house.
-
No training on your data
Your documents are there to answer your questions, not to train models. They never become training data, for anyone.
-
You decide the permissions
Reperix has its own group-based permission system: you assign who sees what, uniformly across documents from different sources. Nobody receives content they don’t have access to in an answer.
-
It searches with the asker’s identity
There is no technical user that sees everything: search starts from the identity of the person asking and filters before searching, not after.
-
When permissions are unclear, nothing is shown
If a document’s permissions are unclear, that document is not shown. Better not to answer than to answer someone without the right.
-
Every customer is isolated
One organization’s data never mixes with another’s. Clear boundaries, by design.
Permissions
The same question, two different answers
Anna and Marco ask the same thing. They get different answers, because the documents they can open are different.
The same question
What does the Rossi S.p.A. agreement say about payment and renewal?
-
Anna
Finance
What Anna can open
- Contratto_Rossi_2024.pdf used
- Allegato_condizioni.pdf used
- Verbale_CdA_marzo.pdf not accessible
- Bozza_rinnovo.pdf permissions unclear
The answer Anna gets
Payment is at 60 days, as stated in the annex. The contract renews automatically on 31/12 unless notice is given.
Sources
- Contratto_Rossi_2024.pdf
- Allegato_condizioni.pdf
-
Marco
Sales
What Marco can open
- Contratto_Rossi_2024.pdf used
- Allegato_condizioni.pdf not accessible
- Verbale_CdA_marzo.pdf not accessible
- Bozza_rinnovo.pdf permissions unclear
The answer Marco gets
The contract renews automatically on 31/12 unless notice is given. On payment, no documents are available.
Sources
- Contratto_Rossi_2024.pdf
The list above is there to explain the mechanism: in his own answer, Marco never learns that documents he cannot open exist.
Deployment
Choose how much control you want
The same Reperix, in three configurations. You move between them without changing product: it’s a deployment choice, not a different piece of software.
-
On-premise hybrid
Everything runs on your own servers.
Your company
- Your documents
- The search across them
What crosses the boundary
Only the few passages relevant to the question leave
The full archive never leaves
Outside your company
- The service that writes the answer
-
On-premise fully local
For the most sensitive contexts.
Your company
- Your documents
- The search across them
- Writing the answer
What crosses the boundary
Nothing leaves, at any point
Outside your company
- No data
-
Cloud / SaaS hybrid
We manage the infrastructure.
Your private area, managed by us
- Your documents
- The search across them
What crosses the boundary
Only the passages needed to answer are sent
The whole archive is never sent
Outside your area
- The service that writes the answer
Your documents are hosted by us, in an area kept separate from every other customer’s.
Security is in the code
Not just promises: how it’s built inside
Principles are worth as much as their implementation. Here’s what actually happens, at the point where it matters.
- The permission filter is applied inside the search, before the results: the system searches only among the documents you can see, it doesn’t search everywhere and then hide.
- Neither permissions nor the asker’s identity leave toward the service that writes the answer: at the point in the code where the request is composed, those fields don’t exist.
- In a hybrid setup only the few passages relevant to the question leave, never the whole archive.
- Connector credentials and shareable-link tokens are encrypted at rest (AES-256), never in clear text in the database.
- The AI provider is configurable: you are not tied to one vendor.
- No “made-up” answers: everything is grounded in real documents and cited.
FAQ
The security questions we hear most
Are my documents used to train the AI?
No. Your documents are there to answer your questions, not to train models. In a hybrid setup, the service that writes the answer receives only the few passages needed for that question; in the fully local setup nothing leaves at all.
Does my data leave my network?
It depends on the configuration you choose. Fully local: nothing leaves, at any point. Hybrid (on-premise or cloud): only the few passages relevant to the question leave, never the whole archive, and never the permissions or the identity of the person asking.
Whose identity does the system search with?
The identity of the person asking. There is no technical user with access to everything: the permission filter is applied before it even searches, so the search happens only among the documents that person can see.
If I change permissions, how soon do they take effect?
From the next question. If you remove someone from a group or change a document’s visibility, the new rule applies immediately: permissions are re-read on every question, not cached once and for all.
How do you keep different customers’ data separate?
Each organization has its own isolated space: one customer’s data never mixes with another’s, and no search crosses the boundary between two customers. It’s separation by design, not a setting to remember.
What happens if a document’s permissions are unclear?
The document is not shown. When in doubt, no: better an incomplete answer than an answer to someone who wasn’t entitled to it. And the answer doesn’t even hint that the document exists.
Where are connector credentials stored?
Encrypted at rest (AES-256). Provider keys, connector passwords and shareable-link tokens are never in clear text in the database, and are never returned to a client.
Have specific security requirements?
Tell us your context: together we’ll find the right deployment configuration.