Transparency & security
See-through, but not leaky.
CompanyWizard shows you what happens in the background — from PII redaction through knowledge retrieval to the model and tool calls — and secures the content at the same time. You see every step, yet nothing comes out that should not.
Transparency and security are seen as opposites
Transparency wants to show, security wants to protect. Reveal everything and you create attack surface; lock everything down and you create a black box you have to trust. For an AI assistant that works with company knowledge and personal data, this apparent contradiction weighs especially heavily.
The conflict disappears once you distinguish: transparency is about the process, security is about the secrets. CompanyWizard makes visible that and how data is processed, without exposing the sensitive content itself.
A visible process, protected content
Every chat request triggers not just a model call but a whole chain of steps. In many tools it runs invisibly — in CompanyWizard it is written into the activity sidebar in real time. You see that the PII pipeline ran, which knowledge sources were drawn on, and which model answered with which tools.
Behind it sits a security layer of role-based access (RBAC), the PII pipeline, user-scoped OAuth and LDAP sign-in. On request, the desktop app keeps everything local, and every installation is strictly single-tenant. Precisely because the content is protected, the process can be shown without risk.
The visible processing pipeline
-
Input redaction
PII in your message is detected and masked. What becomes visible is that the pipeline ran — not the plain-text data.
-
Knowledge retrieval (RAG)
The knowledge base is searched semantically; you see which content was drawn on.
-
Result redaction
PII in the search hits is handled too, before they go further.
-
Model call
The request goes to the chosen model — you see which one.
-
Tool calls
If the model uses tools such as SQL or mail, it becomes visible which ones were active and to what end.
What this means for you
- Trust through observation rather than assertion — you watch the PII pipeline run.
- Cost and usage statistics per model and per user, based on real token feedback.
- Full control over import and export: nothing enters the knowledge base unseen, nothing leaves it unchecked.
- Security in depth: RBAC, PII pipeline, user-scoped OAuth, LDAP, local operation, single-tenant.
Honestly: the residual tensions
-
RBAC limits transparency on purpose.
Not everyone sees everything, and that is intended. At the access boundary, security takes precedence. What you see is always what you are entitled to — not necessarily everything that exists.
-
Per-user statistics are also surveillance.
A cost breakdown by user can act as performance or behaviour monitoring. In Germany that is a matter for data protection and co-determination — use it deliberately, communicate it, and aggregate where that suffices.
-
A visible process is not exposed content.
The sidebar shows that the PII pipeline ran, not the plain-text data. The quality of protection depends on detection — it does not replace careful classification of sensitive content.
-
Security is only as good as its configuration.
RBAC roles, LDAP and OAuth connections must be set up and maintained cleanly. The biggest residual risk is misconfiguration, not a missing feature.
-
Maximum locality costs model quality.
Keeping documents and model fully local gains data sovereignty but, depending on the local model, means trade-offs in performance and speed. A deliberate trade-off between protection and capability.
Sounds right for your organisation?
In a 30-minute demo we show the concept on your concrete use case — no obligation, with concrete suggestions at the end.