What changes for privacy

The key question is one: where does your data end up?

  • Local AI: your request is processed inside the computer. Nothing leaves, no provider sees it, nothing is stored elsewhere. It's the maximum of confidentiality, by design.
  • Online AI: the request travels to the provider's servers, gets processed there and can be stored (and, depending on the settings, used to improve the models). Even by turning off the use for training, the content still passes through and sits on their systems.

For privacy, then, they're not equal: with local the data doesn't leave; with online it leaves and you trust how it's handled.

Which to choose for your case

  • Sensitive data (clients, contracts, third parties' personal data, business secrets): local. When confidentiality is non-negotiable, it's the only choice that truly shields you.
  • Normal work where quality and convenience matter (brainstorming, public texts, general questions): online. It's more capable, faster to use and doesn't require a powerful computer.
  • Most people: mixed. Keep a local AI for the few confidential tasks and an online AI for the bulk of the work. A common strategy is local for the data that must not leave, online for when you need the highest quality.

A concrete example

An accountant uses AI for two different things. To write the firm's newsletter and answer generic tax questions he uses an online AI: those are texts meant to go out anyway, and the higher quality saves him time. No privacy problem.

To analyze clients' financial statements, instead, he uses a local AI installed on his computer: those numbers must not end up on anyone's servers. It goes a bit slower, but the data stays inside the firm. He realized he didn't have to choose a single AI: he had to use the right one for each type of data. Privacy is decided by the sensitivity of the content, not by habit.

When the trade-off weighs (and how to manage it)

If the local one is too unbrilliant for that task

Free local models are a few months behind the best online ones. If for a confidential job the local one isn't enough, anonymize the data (remove names, exact figures, identifying details) and then you can use it online with less risk. Data stripped of its references means reduced potential harm.

If you think paid online is as private as local

Even paid plans with privacy promises process the data on the provider's servers: it's a different level of trust from "they don't leave at all". For many uses it's perfectly fine to trust, but if the rule is "this data doesn't leave the company", only local respects it to the letter.

If you don't know how to keep the two together

Managing two AIs seems complicated, but in practice it's simple: a mental rule. Before pasting something into an online AI, ask yourself "if this ended up on an external server, would it be a problem?". If yes, use local; if no, online. The one-second question sorts almost everything for you on its own.

A tip from someone who actually uses it

Don't choose "local or online" in the abstract: make a two-line list of what is truly confidential for you (usually a few categories: client data, legal documents, work secrets). Those go local, full stop. Everything else can stay online without worries. Deciding once which are your untouchable data spares you from redoing the reasoning in every conversation.

Frequently asked questions

Is online insecure?

No, "insecure" is the wrong word: the big providers protect the data and offer privacy settings. The point isn't security but control: with online you trust how they handle the data, with local you don't have to trust anyone because it doesn't leave. For many uses trusting is perfectly fine.

Can I use local only for the delicate things and online for the rest?

Yes, and it's the most sensible choice for most people. There's no obligation to choose only one: you install a local AI for confidential data and keep the online subscription for everyday work. You use the right tool depending on the data.

For privacy is local always the best choice?

For privacy in the strict sense yes: it's the only option where the data doesn't leave your computer. But "best in absolute terms" no: you pay in quality, speed and the power demanded of the computer. The right choice balances privacy and practicality, and that's why for many the answer isn't "one or the other" but "both, each in its place".