Local AI for preparing a text copy to share externally

The name is removed, and the story still points to it

Say a company is preparing a description of a problem to discuss with a supplier. The text contains an employee's name, contact details and the name of an internal project. The app proposes replacing the name with a description of the role and removing the contact details. A reviewer notices, though, that a detail in the next paragraph still identifies a project the supplier should not know about.

The employee corrects the copy and checks that the supplier will still understand the problem. Removing a name on its own does not confirm that the text is anonymous or that it may be passed on. The rules the company uses to assess material apply even when the proposed changes come from a model running inside the company.

Proposals shown next to the original

Presidio describes replacing, masking and removing detected passages of personal data. These are functions for processing a given text. In the proposed app the user should see what has been marked and be able to add their own correction. The company's requirements may also cover commercial information that no standard category of personal data describes.

It helps to tell removing a passage apart from replacing it with a neutral description. Removing too much can leave the recipient unable to follow the case. The owner of the material judges both the information left in and the meaning of the whole copy. The finished text should be easy to read again before it is sent.

When a template is enough

If the partner always needs the same fields, a form or a description template may work better. The employee then writes the text from the start within the agreed scope. With a few short documents, editing by hand can be simpler than building a tool.

Local help is worth comparing where descriptions are long and varied, and the team regularly prepares versions of them for different recipients. What counts in the trial is the number of missed passages and the effort spent on corrections, including restoring the meaning of a text that was cut too hard. A smooth final version does not replace reviewing it.

Scope limited to text

The first app can work on selected text documents. Files with images, comments or other layers need a separate agreement on how far they are checked. A visible change in the text gives no grounds to say the whole file is ready to share.

The original and the copy serve different purposes. The team should be able to recognise the version meant for the recipient, while access to the original stays with the right people. If the company requires local processing, it also covers working copies and saved comparisons. Maintenance of the tool and service access need to be agreed with IT.

The proposed work with the team

Syntalith can start from fictional texts that match typical case descriptions. The person responsible for sharing shows what should stay in the copy and how they judge whether it is useful. On this material you can look at the proposed changes and the approval step before the company allows real documents.

After the trial we agree on the type of texts covered, the recipients and how the checked copy is kept. The scope can include a connection to your current document flow, with no automatic sending. This is no service that certifies anonymisation or the legal compliance of the material.

Describe a text the team regularly prepares for a partner. A fictional example and the purpose of sharing are enough. We price a private model after a call about the data and the hardware, and an app on it starts from €3,000 net (pricing).

The proposed AI app can show the original next to the copy for checking. The company decides what information may be shared, and the decision to pass it on stays with the named person.

What preparing copies with local AI costs and when it pays back

Where to start

In the free process review we choose one task and 20 to 100 documents the team has already worked on, and within 2 working days you get a written summary with a price. We roll out the local model as part of private LLMs, and the app around it as an AI app.