The short answer
Yes, and more easily than most people expect. A recent laptop with plenty of memory can run a small model well enough for drafting, summarising and questions about your own files. It will not match the best hosted models, and it will make your fans work. The reason to do it is privacy, not performance.
What decides whether it will work
Memory, mostly. A model has to fit in memory to run at a sensible speed, and that is the wall people hit first.
Smaller models are published in compressed forms that trade a little quality for a lot of size. That compression is what makes ordinary hardware viable at all.
A graphics card helps enormously, and recent Apple machines do well because their memory is shared with the graphics processor. An older laptop with modest memory will technically run something, slowly, and you will not enjoy it.
PIXEL / THE COACH
The honest test is whether you would still use it on a Wednesday. Plenty of people install one, admire it for an evening, and quietly go back to the website.
What it is genuinely good for
Anything you would rather not send anywhere. Medical letters, legal documents, client material, a diary, anything with other people's details in it. The text never leaves the machine, which is a promise no policy can match.
It also works with no internet, costs nothing per use once installed, and cannot be discontinued, rate-limited or repriced underneath you.
Roughly what to expect
- A recent laptop, lots of memory
- A small model, useful for drafting, summarising and questions about your files.
- A gaming machine with a good card
- A mid-sized model, noticeably better, fast enough to feel normal.
- An older or low-memory laptop
- It will run something. You will not want to use it twice.
The shape of what is realistic rather than a benchmark. Speed varies enormously with the model and the machine.
Be honest about the gap
A model you can run at home is not the model behind the best hosted products. It will be weaker at long documents, careful reasoning and anything needing breadth of knowledge.
For many everyday jobs that is fine. For the hard ones you will notice, and pretending otherwise leads to a disappointing evening and an unfair conclusion about the whole idea.
The sensible arrangement is both: local for anything private, hosted for anything hard and not sensitive.
What to actually do
- Check your memory first. That is the number that decides what is possible.
- Start with a small model and a real task you do weekly, not a benchmark.
- Use it for the private jobs and keep a hosted tool for the difficult ones.
See it for yourself
Reading about it only gets you so far. This one is written to show you the thing described above, rather than describe it again.
My laptop has sixteen gigabytes of memory and no separate graphics card. What could I realistically run on it, and what would be noticeably worse than using the website?
Then paste it into whichever assistant you use — ChatGPT, Claude, Gemini or Copilot. Not sure which? Meet them all.
If you remember nothing else
- Memory is the limiting factor, and compressed models make ordinary hardware viable.
- The text never leaves your machine, which no privacy policy can promise.
- It works offline, costs nothing per use, and cannot be withdrawn or repriced.
- It will be weaker than the best hosted models on hard work.
- Local for private material, hosted for difficult material, is a sensible split.
Where this came from
- Ollama, running models locallychecked 26 September 2026
Anything here that turns out to be wrong is worth telling us about. Figures that move — prices, sizes, energy use — are kept in the registry with their dates, rather than written into the prose where they would quietly go stale.
Questions about “Running AI on your own computer”Asked by readers, answered by us. Open to read them.
Open to load them.
Ask about this page and we will answer it here, where the next person can read it too.
Nobody else can answer you, and you cannot answer anybody else. This is not a forum — we would rather say that than run a room we cannot watch at three in the morning.
Your question goes up with the answer and without your name. Nothing identifies you, ever.
If we cannot answer it we will tell you why, and that one stays between us.