Local AI or online AI: what happens to your document?

"Uses AI" can mean two very different things for your document. In one case the text travels to someone else's servers; in the other it never leaves your computer. Knowing which one you are dealing with is the first privacy question to ask.

With online AI, the document travels

When you paste text into a chatbot, or upload a file to a web service that "uses AI", the content is sent over the internet to the provider's servers. That is where the model runs. The answer comes back to you, but what you sent has been processed, and often kept for a time, outside your organization.

What happens next depends on the provider, your plan and your settings: how long conversations are kept, whether staff at the provider may look at them, and whether they can be used to improve future models. These rules differ from one service to another and change over time. They are policies you have to read and trust, not something you can check from your desk.

None of this makes online AI bad. It is often the most capable option. It simply means that the document, and every name and number in it, has left your hands.

With local AI, the document stays put

Local AI (also called on-device AI) means the model is installed on your computer, like any other program. The document is opened, read and analysed there. No text is sent to a server to be processed, so there is nothing for anyone else to store, look at or reuse.

The trade-off is real: a local model has to run with the memory and processor of an ordinary computer, so it is smaller and more specialised than the largest online models. For a focused task, such as finding names, addresses and account numbers in a document, that is usually a good fit.

Questions to ask about any tool that "uses AI"

• Where does the AI run: on my computer, or on the vendor's servers?
• Is anything sent while I work: the document, the text found in it, file names, or fingerprints of them?
• Does it keep working with the internet unplugged? If not, what does it need the connection for?
• Are models downloaded while I work, or installed with the software?
• Does it learn from my documents? If so, where is what it learned stored, and who can see it?

A clear vendor answers these in writing. "We take privacy seriously" is not an answer.

Using both, safely

Many teams will keep using online AI for summaries and drafts. The safer order is simple:

1. Remove the identifying details locally first: names, addresses, numbers, anything that points to a person.
2. Review the cleaned copy, page by page.
3. Only then paste or upload the cleaned copy to the online tool.
4. Check your organization's policy. Removing identifiers reduces risk, but it does not by itself decide what you are allowed to share.

For a step-by-step approach, see how to clean documents before using ChatGPT or Claude.

Where Anonyma stands

Anonyma uses local AI only. Its models are installed with the software, and no model is downloaded while you work. It reads your PDFs, scans, Word, Excel and PowerPoint files, e-mails and images on your Windows computer, and nothing is uploaded. Anonyma also learns from your choices, and what it learns stays on your computer. The internet is used only to check your licence and look for updates, never to send document content.

Automatic detection can still miss something, so you review every page before exporting, and you remain responsible for what you share.