Explainer · 03
What can it actually do?
A private AI machine does the everyday work people use ChatGPT for — writing, summarising, answering questions — without sending anything out. It's very good at most of it. It isn't the best in the world at all of it, and it's worth knowing which is which before you buy.
What it does well
-
Writing and editing
Drafting emails, letters, reports and proposals. Rewriting for tone, tightening, and translating between common languages.
-
Summarising
Turning a long contract, transcript or report into the points that matter, with the original never leaving your office.
-
Searching your own documents
Ask questions across your files — “what did we agree with this client about payment terms?” — and get an answer that points to where it came from.
-
Coding help
Explaining, writing and reviewing code, including code you couldn't paste into a cloud service.
-
Images
Generating and editing images locally, on machines built for it.
Where cloud models are still ahead
- The hardest reasoning. Long, multi-step problems in maths, science and research are where the largest cloud models pull away.
- Very long documents at once. Cloud models can hold more text in a single conversation. Locally, very large collections are searched in pieces rather than read end to end.
- Live information. A local model knows what it was trained on and what's on the machine. It doesn't browse the web unless you set that up — which then means reaching the internet.
What it feels like to use
You open a window and type, the same as any chat app. Answers appear as they're written. There's no queue, no usage cap, and no message telling you to come back in a few hours.
Bigger machines run bigger models
How capable the AI is depends mostly on how much memory the graphics card has — more memory fits a larger, smarter model. That's the main thing that separates the tiers, and why we pick hardware around the work you actually do rather than the most expensive part on the shelf.