What Is Local AI? A Plain-Language Explanation for Business Owners

"Local AI" (you'll also see on-premise AI or private AI — same idea) means the AI model runs on a computer you own, in a building you control. Your staff still get a chat window that drafts, summarizes, and answers questions. The difference is invisible on screen and enormous underneath: the thinking happens on your hardware, and the material you type in never travels beyond your own network.

The Mental Model

When someone at your office uses ChatGPT, every prompt — including whatever client material they pasted in — is sent over the internet to a provider's data centre, processed there, and sent back. The provider's terms govern what happens to it.

Local AI removes the trip:

• Your staff type into a chat page served from inside your network
• Your network carries the request down the hall, not across the internet
• Your hardware runs the model and sends the answer back

That's the custody chain — three hops, all yours. There is no fourth hop.

What It's Made Of

Three pieces, all of which have matured fast:

1. An open model — the "brain." Downloadable AI models released for anyone to run (from Hugging Face and similar repositories). They come in sizes; bigger is smarter and needs stronger hardware.
2. A runtime — software like Ollama that loads the model and runs it on your machine.
3. An interface — a chat page like Open WebUI that your staff open in their browser, exactly like any web tool, except the address is inside your own network.

The hardware for all of this is a single capable machine for a small team — a workstation with a strong graphics card, roughly a lunchbox-to-shoebox amount of space.

What It Honestly Can't Do

An honest guide states the limits. Local models lag the largest cloud models on frontier reasoning. They don't automatically know current events. A local setup is also yours to keep running — updates, backups, and the occasional restart are now your (or your IT support's) job, the same as any office system. And a badly chosen model on underpowered hardware is genuinely frustrating — matching the setup to the workload is most of the craft.

Worked Example

Illustration — a fictional office, showing the pattern.

A three-lawyer firm installs one workstation-class machine in their office. It runs an open model behind a chat page reachable only on the office network. The lawyers use it to draft contract clauses, summarize discovery documents, and rewrite client letters in plain language. The firm's confidentiality position is unchanged from the pre-AI days: client files live on office systems, period. Nothing new leaves the building, because nothing leaves the building.

Where to Go Next

If the mental model makes sense, the next questions are practical: what does it cost, and is your office actually ready — which is what the readiness quiz estimates in five minutes.