Making network folders and file servers searchable

A file server becomes searchable when an AI knowledge base connects the shared network folders, reads the content of the files and turns scanned documents into text using text recognition (OCR). After that, your team asks in plain language – “How did we set up the dosing system at the dairy?” – and gets an answer together with a reference to the original file. The server itself stays untouched: same folders, same files, same way of working, only now things can finally be found.

In brief

• Most file servers hold the collected knowledge of ten or fifteen years – but it can only be found by those who still know where it is.
• Windows search finds file names, but not the content of scanned documents, and it does not answer questions.
• The connection reads the existing files, turns scans into text using text recognition (OCR) and keeps the index up to date continuously – without changing your filing.
• A big clear-out beforehand is not necessary; only clearly private, technical or confidential areas are excluded.
• Permissions are set before the connection: anyone who may not open a folder may not see its content through the search either.

Why can nobody find anything on the file server any more?

In many companies, the file server is the oldest digital memory: project folders going back to 2008, quotes, inspection reports, photos, operating manuals, plus generations of file names like “Quote_new_final_v2_corrected”. The structure has grown over the years, every employee has built their own logic into it, and those who have left the company have taken their logic with them. Every managing director knows the result: the file exists, but nobody can find it.

The built-in Windows search only helps to a limited extent. It finds file names and, with luck, text in Office documents – but it does not answer questions, and it cannot see the content of scanned PDFs at all, because technically a scan is a photo, not text. So anyone searching already has to know what the file is called and roughly where it is. But that is exactly the knowledge that is missing after a few years.

How does connecting a network folder work?

When connecting, the knowledge base links up with the server’s shared folders and works through the existing files once: it reads the content of the files, converts scanned documents using text recognition (OCR – turning an image into searchable text) and builds a searchable index from them, comparable to the index of a book. In doing so, the system picks out the relevant attributes for each document from its content – customer, project, installation, date, topic – and fills them in automatically, regardless of which folder the file is in. Only the document type is suggested by the system; it is confirmed by a person in the review area, because it determines who gets to see a document through the search – for unambiguous documents this is done quickly.

After that, the connection remains in place: new and changed files are picked up continuously, and the index stays current. Important in practice: the server is only read. Nothing is moved, nothing is renamed, nothing is deleted. Your team keeps working exactly as before – the difference only shows when searching.

What can AI search do that Windows search cannot?

The difference lies less in speed than in the kind of question you can ask:

Situation	Windows search	AI search in the knowledge base
File name known	Finds the file	Finds the file
Only content known (“it was about the dosing pump”)	Finds Office files with the keyword, not scans	Also finds scans, thanks to text recognition
Question instead of keyword (“How was the system set up?”)	No answer	Answer in full sentences with a source
Different word from the document (“cooling unit” instead of “refrigeration system”)	No match	Understands the meaning and finds the document anyway
Knowledge spread across several files	Open each file individually	Summarises and names all sources

The source is not an extra but a principle: every answer points to the original file on the server, because AI can be wrong – the check is always made against the document. Solutions like the AI knowledge base from Goma-it are designed for exactly this: answers only from your own documents, every statement backed up, data stored on EU servers and no use of your content to train third-party AI models.

Do you have to tidy up the file server first?

The reassuring answer: no, at least not thoroughly. The widespread idea that you first have to spend months sorting folders before an AI search makes sense is the most common reason why such projects never start – and it is not true. The search finds content even in structures that have grown over time, and duplicate detection ensures that five copies of the same quote do not show up five times.

What makes sense is a rough sort in an hour rather than a big clear-out over a quarter: folders with private files, software archives, old full backups and clearly dead material are excluded from the connection. One limitation should be stated honestly: where the server holds contradictory content – say, three versions of a work instruction from three different years – even the best search cannot know which one applies today. Such cases only really come to light through the new findability and can then be cleaned up in a targeted way: mark the valid version, archive the old ones.

How are permissions and confidentiality preserved?

A file server almost always contains areas that are not meant for everyone: HR records, cost calculations, management folders. Before the connection, therefore, one simple question needs answering: who may find what through the search? The knowledge base’s role management reflects this – confidential areas are either not connected at all or restricted to a defined group of people.

The order is crucial: set permissions first, then connect. A search that accidentally makes salary lists findable for everyone damages trust in the whole system – and it is exactly this trust that is needed for your team to adopt the new search. When in doubt: start restrictively and widen access deliberately, not the other way round.

What belongs on the file server – and what in the ERP?

For all the new findability, an honest classification is worthwhile: documents such as invoices, delivery notes and orders have their leading system in the ERP and accounting – that is where they are recorded, invoiced and retained in compliance with the law, and the knowledge base changes nothing about that. The ERP knows what happened. The file server, by contrast, typically contains what no ERP covers: the how. Setup records, project documentation, instructions, correspondence, photos.

This knowledge is the real treasure that the connection brings to light. The question “What did we invoice the dairy in 2021?” is answered by the ERP. The question “How did we configure the system back then, and what do we need to watch out for when servicing it?” will in future be answered by the knowledge base – with the setup record from the file server as the source.

How do you actually get started?

The proven way in is a single, valuable area rather than the whole server: for example, the project folders of the last five years or the folder with instructions and records. This quickly shows what the search can do, which permissions need tightening and where contradictory old material is stored. After that, the scope is extended area by area. The file server is usually the richest source, but not the only one – how it fits together with email mailboxes, SharePoint and paper archives into an overall picture is shown in the overview Connecting documents and systems.

If you want to know what such a connection would look like for you in concrete terms – from the first look at your folder structure to day-to-day operation – you will find the typical project process under Process of the rollout.