Local AI for factory manuals without internet

The operator needs a page they can read

A machine of a specific version runs at the workstation. The person responsible for documentation approved its manual, and a PDF copy sits on the computer. The operator remembers that the document contains a table of codes, and does not know the chapter name. A good answer points to the manual for this version of the machine and opens the page with the table in the local file. A summary of the table or a link to the manufacturer's portal, which cannot be reached, leaves them without the document.

Change one detail: the local collection has no manual for this version. The assistant shows that gap. A similar document gives no basis for presenting the page found as the right one. The person responsible for documentation decides which material to make available.

What a local model for factory manuals costs and when it pays back

The count covers the time spent searching, reading and compiling the material. What stays is the time to check the output against the source, 25% of today's work in the scenario. We measure both times in a trial on 20 to 100 cases the team has already handled.

What really works without internet

A local model runs on a computer in the plant, while the files may sit in an outside portal, the search may use the cloud and the sign-in may need internet. Microsoft, in its description of the Foundry Local architecture, states that downloading the model and components needs a network, and once downloaded the model can work offline from the local disk. The chat may answer without internet while opening the PDF still fails, so the trial goes the whole way to reading the page. The operator also sees which version of the manual they are using and how long the offline copy may be used.

When searching the PDF is enough

If operators use a few well-described manuals and know the codes they look for, local files with ordinary search solve the problem. A model is worth comparing when people describe the information they need in their own words and miss the author's term. We compare on the same work: the operator looks for the page the usual way, then with the assistant, and judges whether they found the right document. The wider subject of authoritative sources is covered in the article on a knowledge base for maintenance.

Where to start

The first trial covers the plant's hardware and a chosen set of manuals. We also trace one document change: from approval of a new manual to opening it at the workstation. We roll out local models as part of private LLMs, and we build the view for working with sources as an AI app. For the free process review, a description of one case and the places where the material sits is enough. Confidential files stay in the company, and you get a written plan within 2 working days.