An AI knowledge base only works if you trust the process. That is why we follow this one.
Before we access your data, we define together:
Only once that is clear do we ask for access.
We follow the least-access principle, in this order:
| Priority | Method | Description |
|---|---|---|
| 1 | Export provided by you | You give us exactly what we need |
| 2 | Restricted folder | Only the files that are in scope |
| 3 | Shared working copy | A copy of your data, never the live system |
| 4 | Time-limited access | Access to one tool, for a defined period |
| 5 | Full access | Only with explicit approval |
We always work on a copy. We never modify your live system.
Every data source is labeled before processing:
| Level | Description | Example |
|---|---|---|
| Public | Already public or safe to reuse | Press releases, blog articles |
| Internal | Business information, safe within the team | Internal processes, team structures |
| Confidential | Private client, financial, or strategy information | Client lists, pricing calculations, contract drafts |
| Strictly confidential | Personal, legal, or medical data | HR files, patient data, passwords, payment information |
Strictly confidential data is excluded by default, unless it is explicitly required and approved.
Every important entry in your knowledge base carries a source annotation:
| Status | Meaning |
|---|---|
| Extracted | Taken directly from a source |
| Inferred | Reasonable assumption, requires approval |
| Confirmed | Reviewed and approved by a human |
| Archived | Kept for history only, no longer active |
AI and automation only make proposals. A human approves before anything becomes trusted knowledge.
The moment an unverified AI recommendation lands as a fact in the system is the moment the system reproduces exactly the problem it was supposed to solve. That is why we never allow it.