Every prompt you type into ChatGPT, Claude or Gemini leaves your computer and lands on someone else's server. For most things that's fine. But when it's a client contract, a student's personal statement, or the numbers behind your business, it's worth knowing there's another option: running an AI model entirely on your own laptop, offline, for free. In 2026 it's genuinely easy — here's how I set it up and what I actually use it for.
Why Run AI Locally at All?
Three reasons, in order of how often they matter to me. Privacy: nothing you type leaves your machine, so you can paste in sensitive documents without reading a single privacy policy. Cost: once it's installed there's no subscription and no usage cap — you can run a thousand product descriptions overnight and pay nothing but electricity. Offline access: it works on a train, on a plane, or when your Wi-Fi gives up during a deadline.
The honest trade-off: local models are smaller than the frontier models behind the big chat apps. They're excellent at summarising, rewriting, drafting and sorting, but noticeably weaker at long, complex reasoning. Think of a local model as a very capable assistant, not your strategist.
What You Need Before You Start
You don't need a gaming PC. A reasonably recent laptop handles small models well. As a rough guide: 8GB of RAM runs the smallest models comfortably, 16GB is the sweet spot for everyday use, and 32GB or more lets you run the larger, smarter ones. Apple Silicon Macs are particularly good at this because their memory is shared between the processor and graphics chip. You'll also want 10–20GB of free disk space, since each model is a few gigabytes.
The Two Easiest Tools: LM Studio and Ollama
LM Studio is where I'd start anyone non-technical. It's a normal desktop app for Mac, Windows and Linux: you open it, search for a model, click download, and get a chat window that looks just like ChatGPT. It even tells you which models will fit on your machine before you download them.
Ollama is the lighter, more powerful option if you're comfortable typing one command. Install it, then run something like ollama run llama3.2 in your terminal and you're chatting within minutes. The real advantage is that Ollama runs a local server in the background, so other apps and automations can talk to it — that's where the business uses open up.
For models, start small and work up. Open-weight families like Llama, Gemma, Mistral and Qwen all come in small sizes (roughly 3–8 billion parameters) that run well on a normal laptop. Download two, give them the same task, and keep whichever one writes more like you want.
Five Things I Actually Use a Local Model For
1. Summarising sensitive documents. Contracts, bank statements, anything with names and numbers in it. Paste it in, ask for the key points and any deadlines, and nothing leaves the laptop.
2. Bulk rewriting. Rewriting fifty product listings or turning bullet points into email drafts is repetitive work that a small model does perfectly well — and with no usage cap, volume isn't a problem.
3. Sorting and tagging. Customer messages, survey answers, or a messy spreadsheet column. Local models are surprisingly good at "put each of these into one of five categories."
4. First drafts on the move. I draft blog outlines and lesson plans offline on the train, then polish the best ones with a bigger model later.
5. Private journaling and thinking out loud. Some things you just don't want sitting in a cloud chat history.
Here's a prompt I use for the first one — small models work best when you're very specific about the output:
[paste document]
Making It Work for Your Business
The real payoff comes when you stop treating the local model as a chat window and start treating it as a free engine behind your workflows. Because Ollama exposes a local API, automation tools like n8n (which you can self-host) can send it tasks on a schedule — categorise yesterday's customer emails, draft replies to common questions, or clean up an export from your store — all without per-request fees. A sensible setup is a hybrid: local models for high-volume, private or repetitive jobs, and a paid frontier model for the hard thinking.
Setting that up properly — choosing which jobs go where and wiring it together so it runs without you — is exactly the kind of thing I build for clients through Tafuma Web Services.
Start Today in Fifteen Minutes
Download LM Studio, pick one small model it says will fit your machine, and give it one real task from your week — a document to summarise or a batch of text to rewrite. Compare the result with what you'd get from ChatGPT. You'll quickly get a feel for what local AI is good at, and you'll have a private, free, always-available assistant sitting on your laptop from now on.