For people who use ChatGPT, or want to
A complete computer with a private AI assistant built in. Browse, write, work, ask. Your data never leaves the room. No subscription. No cloud. Yours forever.
Most people who use AI today pay around £20 a month for ChatGPT (or Microsoft Copilot, or Claude, or Gemini). Their conversations live on someone else's servers in another country. Their data is the product.
The Arsenale Appliance is a small computer that you buy once. It has a real AI assistant built in: the kind that can answer questions, write, plan, summarise documents, do research, help you code. The assistant runs on the computer itself. Your conversations stay on your device. There is no monthly fee.
It is also a complete computer. You browse the web, send email, watch videos, work in spreadsheets, and use the AI any time you want, all on the same machine. Plug it in, set it up, use it.
Concrete, on-the-machine, no-internet-required-for-the-AI-part:
Like ChatGPT, but running on your device. No round trip, no rate limit, no monthly token cap.
Emails, documents, posts, scripts, code. The assistant helps you draft, edit, reformat, tighten.
Drop in a research paper, a contract, a meeting transcript. Get the key points back in seconds.
Browse the web normally. Ask the assistant to read pages, compare sources, build a brief.
Trip itineraries, project plans, weekly schedules. Talk through it; the assistant keeps the notes.
Major language pairs, all locally. No third party reading what you sent.
Write scripts, debug, learn a new language, glue tools together. Same kind of help a paid AI gives you, no metering.
Email triage, customer-message drafts, light bookkeeping help, recurring admin work. Set up once, runs daily.
An honest, head-to-head look at the choices a person currently has if they want a computer and an AI assistant.
| Arsenale | Windows PC + ChatGPT Plus |
Cloud-only AI (no PC bundled) |
|
|---|---|---|---|
| Up-front cost | £2,499 once | ~£800 PC | £0 |
| Required monthly fee | £0 | ~£20 ChatGPT + ~£7 Microsoft 365 | ~£20 (Plus) or £155+ (Pro) |
| Where your conversations live | On the device. They never leave. | Stored by OpenAI, on their servers, in another country. | Stored by the provider, on their servers. |
| Token limit / usage cap | None. Use it all day. | Plus tier rate-limited; Pro tier £155+/month. | Plus tier rate-limited; Pro tier £155+/month. |
| Internet required for the AI | No. Browse needs it; the AI does not. | Yes. Always. | Yes. Always. |
| What you own at year five | The unit. Repurposable. Trade-in eligible. | A five-year-old PC. Zero accumulated AI. | Nothing. You stop paying, you stop having it. |
| If the provider changes price or policy | Doesn't affect you. You own it. | You renegotiate or switch. | You renegotiate or switch. |
| Sensitive documents | Process them. Nothing leaves the room. | Don't paste them. Conversations may be retained. | Don't paste them. Conversations may be retained. |
The pure-pound comparison comes out roughly even. The difference is what you have at the end of five years.
| Arsenale | Windows PC + ChatGPT Plus + Microsoft 365 |
|
|---|---|---|
| Up-front | £2,499 (one time) | ~£800 PC + ~£100 Windows licence |
| Monthly recurring | £0 (no required subscription) | ~£27/mo (ChatGPT Plus + Microsoft 365 Personal) |
| Five-year total | ~£2,499 | ~£2,520 |
| What you have at year five | The appliance, refurbishable, with trade-in value | A five-year-old PC + zero accumulated AI |
| If you use ChatGPT Pro instead | ~£2,499 | ~£10,200 over five years |
The headline: at the same five-year spend, you can rent the AI from someone else, or own one. We think owning it is the better trade for the same money.
The appliance runs Syntex OS, a Linux-based operating system designed around the AI assistant. It will not run Windows .exe files directly. It runs all major web apps (Microsoft 365 web, Google Docs, Notion, Figma, Slack, anything browser-based) and standard Linux software. Native Syntex OS productivity tools are on the roadmap: LLM-first equivalents to Excel, Word, and presentation software, designed to talk to the assistant rather than to be talked to.
Around 18 tokens per second. For comparison: ChatGPT Plus serves replies at roughly 30–50 tokens per second. So the assistant on Arsenale is somewhat slower per second, but it has no rate limit, no daily cap, no surcharge for heavy use, and no queue when the provider is busy.
A 120-billion-parameter mixture-of-experts model: the same class that powers commercial cloud AI services. The weights live on the appliance, not in a datacenter.
We support it. Five years of software security updates are included in the purchase price (no annual fee for security). For hardware faults during the warranty period, we replace or repair.
The appliance keeps working. Operation does not depend on our continued existence; there is no cloud back-channel, no kill switch, no licence-server dependency. The hardware and the software you've paid for remain yours and remain functional.
No. Plug it in, walk through a four-step first-boot, start using it. No command line, no configuration files. If you can use a normal computer, you can use this.
Yes, it is most of the point. Documents you would not paste into a cloud AI (medical records, contracts, intellectual property, customer data, legal correspondence, anything under NDA) can be processed here without leaving the device.
For most everyday tasks (drafting, summarising, research, coding, planning, translation), yes, it is competitive. The largest cloud frontier models are slightly more capable on the most demanding reasoning problems. For the work most people actually do, the difference is not noticeable.
The AI works fully offline. Browsing, email, and anything web-based needs internet, the same as any other computer. The point of the appliance is that the AI doesn't need to phone home.
If you depend on Windows-only desktop software (specific games, specific business tools that have no web equivalent, specialist creative suites that are Windows or macOS only), this is not yet a full replacement; it is a great second machine. If you need the absolute frontier of cloud reasoning capability for research-grade tasks, the largest cloud models still have an edge there.
Reserve a unit at the launch price, or write to us with the questions this page didn't answer. Either is fine.
Want the technical detail underneath? See the platform overview and the trust page.