The approaches at a glance
| Approach | Strength | Limit |
|---|---|---|
| Cloud AI with a contract | Full model quality. | The originals sit with the provider. |
| A purely local AI box | The data stays in the house. | Weaker than frontier models. |
| Prompt filters (browser, desktop) | Simple and cheap. | Per text, often manual, no consistency across an archive. |
| Hosted pseudonymisation gateway | Runs automatically. | The mapping table sits with the service provider. |
| IT-InterMirror | Archive shadow copy and routing switch (in development), key with the customer. | Version 0.1 Alpha. Languages: see Measurement. |
What IT-InterMirror sets out to do differently
- A lasting shadow copy of the whole archive, with the same placeholders in every document. Status: in development.
- A routing switch: a local AI answers simple questions, and only heavy tasks go to the cloud as a shadow copy. Status: in development.
- Operation on hardware at the customer’s site. The key stays there.
According to our own research (as of November 2026, without guarantee) we know of no solution that combines all three points.
Which approach fits?
That depends on your documents, your language and your duties. For a single non-critical text, a simple filter is often enough. For a whole archive with many names and numbers, an approach that stays consistent across all documents pays off. We will also tell you when IT-InterMirror is not the right choice.
Interested?
We are looking for partners. We test IT-InterMirror with made-up or cleared examples first. Only then does real text go through.
This page is also available as Markdown for AI reading systems. /en/approaches.md