Why on-premises AI

An assistant that answers questions about company documents reads everything: procedures, drawings, tenders, contracts, sometimes personnel records. If that text leaves the company, every permission you built over twenty years of file servers stops counting.

The problem is not the model, it is the path

Most assistants work like this: the document is uploaded to an external service, indexed there, and answers come back. The model can be excellent, but the path of the data is clear: it leaves the network, goes through a vendor and stays in an index you do not control.

For a small company this means three concrete things:

• customer security questionnaires ask whether the vendor can access the data, and "only if it is inside your network" is not accepted;
• records of processing and impact assessments get harder, not easier;
• an unauthorised data egress is an incident to report, even when the vendor is careful.

What changes with borc

borc runs on a machine inside your network. Documents are read from the sources you configure (network shares, SharePoint, upload collections), indexed into a local Postgres, and never copied anywhere else.

Chat keeps working without internet. The only outbound connections are the updater's, towards platform.borc.ai for licence, version and numeric metrics: no content, no file names, no users.

The sentence to use, and the one to avoid

The correct sentence is: your data and your users never pass through our systems, and nobody at borc can get in without your authorisation. That is not the same as "not even we can see them": whoever administers the machine has technical access to the database, like any technician with access to the server, and it is right to tell employees so.

One trust point cannot be removed: we write and update the software. The countermeasures are verifiable: an offline signing key, images pinned by digest, no downgrades, and manual approval of updates for those who ask for it.