How to chat with your PDFs without uploading them to the cloud

Every week, millions of people upload sensitive documents to AI services to ask simple questions about them. A contract they want to understand. A medical report they want explained in plain language. A financial statement they need to summarise. The AI answers well. But the document — and everything in it — has now left their computer and landed on a server they don't control.

Most people assume this is simply the price of using AI. It isn't. Local AI has reached a point where you can get the same quality of answers from documents on your own computer, with nothing leaving it. This article explains how.

Why cloud AI document tools are a privacy concern

When you upload a document to a cloud-based AI service, you are trusting that service with every word in that document. Depending on the provider's terms, your document may be used to train future models, stored on their servers indefinitely, or accessible to employees for safety review. Even providers with strong privacy policies represent a potential point of exposure — through data breaches, legal requests, or policy changes.

For many documents this is an acceptable trade-off. For others — legal contracts, medical records, financial reports, client data, internal company strategy — it is not. And in some cases, uploading documents to third-party services may violate regulatory requirements such as GDPR, or your organisation's own data handling policies.

What kind of computer do you need?

Running AI locally asks more of your computer than everyday tasks like browsing or email. The good news is that most computers bought in the last few years will work — the experience just varies depending on what's inside yours.

The best experience comes on a Mac with Apple Silicon (the M1, M2, M3 or M4 chips — any Mac sold since late 2020 has one of these), or a Windows or Linux machine with a modern NVIDIA graphics card. On these computers, answers typically arrive within a few seconds.

On older or less powerful computers, Glealu will still work, but responses may take a few minutes each rather than a few seconds. That's fine for occasional use — though it can feel slow if you're using it heavily throughout the day.

Not sure which category your computer falls into? Check your computer's compatibility on the Glealu website — then you know what to expect.

The practical steps

Setting up local document AI used to require technical knowledge and a willingness to work with command-line tools. That has changed. Glealu guides you through the entire setup.

The general process is: install a local AI engine (Glealu uses Ollama, the most widely used engine), download a language model and an embedding model to your machine, then use an application that connects to Ollama and handles the document indexing and retrieval. Once set up, using Glealu feels just as simple as any cloud-based tool — you upload a document, ask questions, and receive answers with source references. The difference is that nothing leaves your machine.

The key difference when using Glealu is that every part of this runs on your computer. You can disconnect from the internet entirely and it keeps working. No account is required after the initial setup. And nothing you upload is ever transmitted anywhere.

What kinds of documents work well?

Glealu currently works well with file types like PDF, txt and md (markup). PDFs are the most common format and work reliably, whether they contain reports, contracts, research papers or technical documentation. Plain text and Markdown files also work well. Scanned documents — images of pages rather than digital text — are more challenging and depend on whether the tool includes OCR capability.

AI tend to perform best when your questions are specific and the answer exists somewhere in the document. "What is the termination clause in section 12?" will produce a better answer than "summarise the entire contract." For complex, multi-step reasoning across very long documents, cloud AI services with larger models may still have an edge — but for the vast majority of practical document questions, local AI is more than capable.

Is it really private?

With Glealu, your documents genuinely never leave your computer. The language model is a static file — it has no network capability and cannot send data anywhere. The embedding model works the same way. The only network connections Glealu makes are for license validation or software updates — never for document content or your questions.

If you want to verify this yourself, the simplest test is to disconnect from the internet — Glealu keeps working without it. For a more technical check, a network monitoring tool such as Wireshark will show zero outbound connections while you process documents and ask questions.