A British software and AI lab · Tunbridge Wells
Software that buys from software.
The interesting part of AI is not conversation. It is transaction — an agent ordering a part, renewing a licence, paying for an API call by the call. URSY builds the software that makes that safe, and proves it on a network of seventy-eight domains first.
A moving illustration of agents settling payments with services. The names, items and amounts are invented to show the shape of a settlement; they are not real transactions and no figure on it is a measurement.
Four questions before any of it is safe
When a machine pays a machine, four things have to be true.
URSY keeps a working thesis on this, written as an argument rather than a pitch and updated with dated revisions. Every build here answers the four questions in code before it ships.
URSY ID gives an agent the same account as the person behind it — one login, one record, across the network.
Scoped, expiring permission — a budget, a category, a date. Anything outside it is refused and the refusal is logged.
By the call, by the item, or not at all — with the price printed before the request, never discovered after it.
A receipt both sides can verify, a stable anchor a machine can cite, and a log that is never edited to make an old claim true.
What gets built here
Three kinds of work, and nothing pretending to be a fourth.
Software built to fit a job that no off-the-shelf product fits. Websites and web applications, built or rebuilt properly rather than dressed up. Applied AI inside a product — something that answers from your own records, with its limits printed underneath it.
Work is taken one engagement at a time, because there is one person doing it.
What has already been built →Usually replacing a spreadsheet that has outgrown itself.
WordPress work happens at plugin and runtime level, not theme level.
Read-only, documented, and keyless where that is safe.
llms.txt, a documented endpoint an assistant can call, stable anchors so a machine can quote a paragraph.
Answers from your own catalogue, names its sources, says when it does not know.
On something already built, written down, faults ranked, reasoning shown.
The network is the laboratory
Seventy-eight domains on one runtime. Twenty-two live. The rest say so.
The register includes public products and holding pages. Figures retain the source property's dated count. Membership was verified on 13 September 2026; each product owns its current availability.
Shared runtime access is available across the registered estate. Accounts, checkout, delivery and product features retain their own controls and published availability; a place on the register does not certify every product as finished.
The full register, holding pages included →A scrolling list of all 78 domains on the network with their status and what each one says about itself. The same list is published in full, and searchable, on the network page.
Honest beats impressive
Name the source, date the fact, state the gap in words rather than filling it, and disclose anything paid for.
ursy.domains publishes catalogue prices with the date they were observed, and says on the record that availability, premium status and price are revalidated before payment. It will not put a name in a basket on the strength of a catalogue row. An earlier version guessed. The guess was taken out and the refusal was written into the code.
A missing number and a number that is zero are different facts, and treating them as one is how a dashboard starts lying.
Open systems and open education →A tea shop that is also an instrument
teas.co.uk is the one site where I am both builder and customer.
It sells tea, infusions, coffee and biscuits from Tunbridge Wells. Underneath the storefront sits a full machine-facing surface: an llms.txt, a products feed, a product-truth endpoint that answers by barcode, a Model Context Protocol server card, an RFC 9727 API catalogue, and a stable identifier on every registered heading.
Above it sits an Under the Hood section that publishes the workings — and lists its own mistakes, dated, including the ones still outstanding.
teas.co.uk is live →What AI systems may use, and how to credit it.
Product truth by barcode, source and date attached.
A Model Context Protocol server card an agent can read.
Every endpoint listed, with the document that describes it.
Questions people ask
Answered plainly.
Bespoke software built to fit a job no off-the-shelf product fits, websites and web applications, public read-only APIs over data you already hold, machine-readable surfaces for a site you already have, and applied AI inside a product.
Lee Tucker, the founder. URSY is independent and founder-led; every engagement gets his direct attention from first paragraph to handover.
You send a paragraph. You get an honest written answer. If it is yes, a scope with its limits printed, a build on the same runtime as the network, and a handover with the workings.
Tunbridge Wells, Kent, in England, working across the United Kingdom. Send the paragraph through the form on the contact page — that is the route that reaches us.
Got something you want built?
Describe the thing in a paragraph and send it. You will get an honest answer about whether it is worth building and whether I am the right person to build it. If the answer is no, I will say so, and say why.
Start with a paragraph →Ask about this site, its runtime or its public evidence. Nicola names her sources and says when she does not know — she is not a substitute for the paragraph.