Scope a tool that does one job well
One user, one task, one measure of success, before anything gets built.
Data, privacy & security
Running a model yourself versus using one over the internet: the trade-offs.
5 min readPlain English
In one sentence
Cloud AI runs on a provider’s computers; an open-weights model runs on a machine you control, so your text stays with you but the upkeep is yours.
Think of it like…
Cloud AI is eating at a good restaurant: a skilled chef, no dishes to wash, but your order goes to their kitchen. Open weights is getting the recipe and cooking at home: more effort, a simpler menu, and nothing leaves your house.
With cloud AI, your prompt travels to the provider, the model runs there and the reply comes back. With open weights, the whole trip happens on a machine you control. Getting there takes five steps.
There’s a middle ground: some cloud services host open-weights models for you. Then your text goes to that host, under its terms.
Good at
Bad at
Client letters at a bookkeeping firm. A small model on one office computer summarizes client letters, so nothing leaves the building. A person still checks every summary.
Site reports with no signal. A surveyor drafts reports on a laptop at rural sites, then reviews them back at the office.
Everyday marketing. A design studio sticks with a cloud assistant for campaign ideas and web copy: nothing sensitive, and a bigger model writes better.
Myth vs reality
Myth: “Open weights means open source, so it’s free to use however you like.”
Reality: Open weights means you can download the trained model. The training data and code often aren’t released, and the license may limit commercial use. Read it before you build on it.
Your team wants to summarize confidential client files, and the rule is that nothing leaves the office. Which setup fits?
One user, one task, one measure of success, before anything gets built.
What people may paste, which tools are approved, and what always gets a human check.
Where your text goes, who can see it, and the settings that matter.