No subscriptions. Ever.
No monthly per-seat fee, no per-token metering, no annual renewal that creeps up 20%. You buy the appliance once and the running cost is the electricity it draws.
A frontier-class open-weight model running on hardware you own, on your network, with no subscription and no per-token bill.
Setup AI ownership for your business today!
£4,000 hardware · £1,000 setup · no subscription
Book a two-hour on-site visit with one of our experts. We install it, configure it around how you actually work, and train your team before we leave.
We answer 8am to 8pm, UK time. Speak to an expert before you book if you have questions — booking takes payment up front.
The two most capable hosted models in the world cost between $1.23 and $3.14 per task at maximum effort. Multiply that by a team, by every working day, forever. An Albatross is bought once. After that, a task costs you electricity.
£4,000 for the appliance and £1,000 to have it installed, connected to your systems and your team trained on it. There is no second invoice.
Cost-per-task figures: Artificial Analysis — Cost per Task, weighted average across the Intelligence Index, read 15 August 2026. Claude Opus 5 (max) $2.34, Claude Fable 5 (with fallback) $3.14, GPT-5.6 Sol (max) $1.23.
No monthly per-seat fee, no per-token metering, no annual renewal that creeps up 20%. You buy the appliance once and the running cost is the electricity it draws.
The hardware is your asset — capital expenditure, on your balance sheet, depreciable. The model weights are open source under Apache 2.0. Nobody can revoke either.
The Albatross serves an API on your local network only. It is never exposed to the internet. Client files, contracts, payroll, patient notes — nothing is uploaded to anyone, because there is nobody to upload it to.
Ten people can hammer it all day and the bill does not move. There is no usage anxiety, so people actually use it — which is the entire point.
Broadband outage, provider incident, rate limit, region down. None of it touches you. The model is sat in your office on your switch.
The appliance learns your business through your own documents and your own context, not a generic assistant. Swap in a better open-weight model as they ship — the hardware doesn't expire.
Hosted models get retired, renamed and repriced on someone else's schedule. Yours runs the version you chose, for as long as you choose.
We do not ship you a box and wish you luck. We install it, connect it to your systems, load your documents, train your team, and support it.
Everything below runs locally, on your own hardware, from plain English instructions.
Describe what you want. It writes the code, runs it, and hands you something working — landing pages, booking forms, internal tools, dashboards.
Trained on your products, prices and policies. Hosted by you, answering out of hours, with no per-conversation charge attached to it.
Multi-step work that runs on its own — read the inbox, extract the data, update the sheet, draft the reply, flag the exceptions.
Read incoming invoices, pull out line items and totals, match them against purchase orders, chase what's overdue, and generate your outgoing ones.
Board decks, quotes, pricing models, month-end reports. Describe the output and it produces the file — including the analysis behind it.
Draft from your own templates, compare versions, summarise a 60-page agreement into the five clauses that matter.
Draft replies in your tone of voice, summarise long threads, prep call notes, and follow up without anyone remembering to.
Point it at your file share. Ask questions in plain English and get answers with the source document attached.
Delivery notes, receipts, site photos, handwritten forms, whiteboards. It sees them, not just text.
One SKU. One configuration, apart from the storage you choose. Every unit built and tested in the UK.
Projected throughput ~28 tokens/second single-stream at 4-bit quantisation, derived from 640 GB/s memory bandwidth against a ~16GB model. This is an engineering estimate, not a measured result — we will publish measured figures once the production unit is benchmarked.
No per-seat licence, no per-token metering, no annual renewal. All prices exclude VAT.
The appliance itself. Built and burn-in tested in the UK, imaged with a hardened Linux stack and your model, serialised, UKCA marked and delivered to your door.
We come to you and make it work inside your business, rather than leaving a box on your desk. This is the difference between owning an AI and owning a computer.
A short call about what your business actually does, what's eating the hours, and which storage tier fits your documents. No obligation, no scripted demo.
Each Albatross is assembled and burn-in tested in the UK to our specification, then imaged, serialised and shipped to you. Typically 10–15 working days.
We come to you. It goes on your network, connects to your systems, ingests your documents, and your team gets trained on it. That's the £1,000 setup.
Your team opens the app and works. Nothing to renew, nothing metered. We stay on the end of the phone for support and model updates.
Not at the very top end, and we won't pretend otherwise — the largest hosted models are bigger than anything that fits on a desk. What the Albatross runs is a frontier-class open-weight model that handles the overwhelming majority of real business work: drafting, summarising, extracting, coding, analysis, reading documents and images. For the small number of tasks that genuinely need a frontier model, the app can route that one request to a hosted API if you choose. You decide, per request.
It stays on your premises. The appliance serves an API on your local network and is never exposed to the internet. We do not receive your prompts, your documents or your outputs. There is no telemetry pipeline carrying your content back to us.
Electricity. It draws roughly 450 W under full load and far less when idle — comparable to a desktop PC. There is no software licence, no per-seat fee and no usage metering.
A small team comfortably. It is built for concurrent use, not one person at a time. If you have heavy simultaneous demand we will spec the memory accordingly before you order.
Three-year return-to-base warranty and UK support. We are the manufacturer of record — you deal with us, not a component vendor and not an overseas call centre.
The hardware is a general-purpose inference machine, not a fixed-function product. Open-weight models ship continuously and better ones will run on the same box. We push new models as part of support.
No. It arrives configured. It boots headless, recovers by itself after a power cut, and your team reaches it through a desktop app that finds it on the network automatically.
Yes. Point it at your document share and it answers over your own material, with the source attached. Tell us your corpus size and we'll spec the storage and memory for it.
Tell us what your business does and we'll tell you, honestly, whether an Albatross is worth it for you. No sales script.