Local AI · home · privacy
AI at home, without a mandatory cloud
A language model can answer from your own machine. The first try uses the computer you already have. A mini PC comes later, only if that try is not enough.
- No purchase The 15-minute guide checks Ollama on the PC you already own.
- Editable kWh The selector recalculates electricity with your own tariff.
- October 2026 Indicative ranges. Check the merchant before you buy.
Where to start
- 1
Try it before you buy
Install Ollama and get one answer. A dedicated graphics card is not required for this test.
First local AI in 15 minutes - 2
Choose hardware afterwards
If the current machine cannot hold the model, the selector applies readable rules: RAM, VRAM, noise, budget. It does not crown a best product.
Open the selector - 3
Compare the cost over 3 years
A ChatGPT Plus-style subscription and a machine running at home are not judged on the first month’s price.
Ollama and ChatGPT over 36 months
The guides, the topic hubs and the selector on this English edition follow the same path as the French site. Shop buttons on accessory pages go to Amazon.com. Figures dated October 2026 stay in euros when that is the figure in the comparison.
For voice, Whisper transcription comes after Ollama already answers. For pictures, local ComfyUI / FLUX is an intermediate step, not the first one.
Other doors: a private family assistant, sizing the hardware, data and privacy. The audience is households and a home lab. A company deployment goes through the contact page.
Beginner and intermediate paths
Times are the ones used on the French guides. They are estimates for someone following the page, not a measured lab protocol.
| Path | Indicative time | You already need | It is for |
|---|---|---|---|
| Beginner — first local AI | ~15 min | A Windows, macOS or Linux PC, and admin rights | Check Ollama and a first reply before buying hardware |
| Ollama on Windows (or Linux / macOS) | ~25 min | The first test worked, or the PC is new | A service at startup and the basic terminal commands |
| Family Open WebUI | ~45 min | Ollama already answers locally | A web interface on the home network, with several profiles |
| Pick a first GGUF model | ~20 min | A first chat already works | Avoid a model that is too heavy for your RAM |
| Budget — Ollama vs ChatGPT (3 years) | ~25 min | None for the reading; the selector has an editable kWh rate | Compare a cloud subscription with a machine running at home |
| Intermediate — local images (ComfyUI) | ~2 h | Stable Ollama, and enough GPU or VRAM | Generate images without sending the job to an online service |
| Intermediate — Docker + Open WebUI | ~1 h | Docker Desktop or a Linux engine, Ollama already tested | A repeatable container stack next to other services |
| Home Assistant + Ollama | ~1 h 30 | A Home Assistant instance and a network you control | Cautious home automation without a mandatory cloud model |
| n8n + a local model | ~1 h | Webhooks or an API, and Ollama reachable | Personal automations for mail, files and reminders |
Dedicated hardware: choose AI hardware or the selector, once the table has made the use clear. All guides: the French beginner section.
What 36 months costs in the October 2026 example
The comparison is in euros, dated October 2026, and indicative. ChatGPT Plus is about €20 a month, so about €720 over 36 months. One local example in the same article: a 32 GB mini PC at about €850, plus about €90 of electricity over 36 months (about 25 W, and a kilowatt-hour at about €0.25), which lands near €940. The article states the assumptions. This page does not convert those euros into pounds.
The hardware selector
It helps after you know the use: chat only, Whisper, images, or a machine left on. Five questions, then readable rules for RAM, VRAM, noise and budget. It does not impose a single best product. The screens are in French, the calculation runs in the browser, and it needs JavaScript. Open the selector. If JavaScript is off, start with choosing AI hardware.
- Uses you want (chat, transcription, home automation, images).
- GGUF size you are aiming at (3B, 7B, 13B and so on).
- Household limits: an indicative budget, silence, power use.
- Technical level: native install, Docker, or a home lab.
- A 12- and 36-month cost with an editable kilowatt-hour rate.
Questions this English page is here to answer
Do I have to buy a computer first?
No. The 15-minute guide uses the PC you already have. Buy hardware only if that test is not enough.
Are the tutorials translated?
Yes. This hub, the guides, the selector, the accessories, the price drops and the short legal pages are in English.
Why are the 36-month figures in euros?
They come from the October 2026 comparison and stay in euros here. A pound price would be a conversion, not the figure in the article.
Which shop do the buttons use?
Accessory buttons and the selector on the English site go to amazon.co.uk, tag roboteducatif-21. The French pages still use Amazon.fr, tag manuso06-21. The disclosure says which is which.
Official sources
Versions and commands should be checked on the publisher’s own pages.