Capxul · Status · 24 August 2026
We spent three weeks rebuilding the base of the product. That work is finished, proven, and ready for production. This morning it moved real money for the first time. From now on we are adding capabilities to a finished foundation rather than propping one up.
This is the ordinary shape of building software. You move fast, you try things, and a good share of what you produce is a prototype — something that proves an idea works but was never meant to carry a product. Prototypes accumulate. Each one is reasonable on its own. Together they become a base that cannot take another floor.
We reached that point. Screens existed that looked finished and did nothing behind the glass. Payments could be started and never completed. The honest choice was to keep stacking and hope, or take the base apart and rebuild it properly.
We rebuilt it. It cost about three weeks. The alternative was to spend those weeks later, with customers watching.
“Foundation” is doing real work in that sentence. It is not only payments. It is everything a financial product needs underneath it before a single screen becomes useful.
Signing up, signing in, creating an account.
Sending, settling, and proving it on the blockchain.
Who may spend what, enforced rather than assumed.
Every movement written down and reconcilable.
When something fails, we see it and know where.
What people do, measured, so we can tell what works.
The last two are the ones nobody asks for and everybody needs. Without them you find out a feature is broken because a customer tells you.
Removed more than 40,000 lines — demonstration apps, a half-finished component layer, tests that checked things nobody shipped. The only way to see what was actually load-bearing.
Signing up, signing in, creating an account. The layer everything else stands on, so it went first.
A new payments contract on the blockchain, a complete record of every movement, and a permission system controlling who may spend what.
An organisation paid a person one dollar. The money left the organisation's account, settled publicly on the blockchain, and appeared in the recipient's balance and activity with a receipt. Every step was checked independently rather than taken on trust.
It is one dollar. It is also the first time this has worked end to end, and it is what every other feature was waiting on.
This was always the plan. One supplier retiring a product mid-flight, with no warning, while we were using it, made it urgent instead of eventual.
Nothing in the money path depends on an outside company any more. Submitting payments to the blockchain, and covering our users' transaction fees so nobody needs to hold crypto to use Capxul — both are ours, running on infrastructure we control.
The application and its database. The service that submits payments. The service that sponsors transaction fees. Our own staging environment. And a complete copy of the entire product that runs on a laptop.
Raw access to the blockchain — running a full copy is disproportionately expensive and always will be. And sign-in, for now. When that becomes a constraint we bring it in-house too. It is not hurting us yet.
Two consequences. It costs less. And when something breaks at two in the morning it breaks in a system we can open up and fix, instead of one where we file a ticket and wait.
We can now run Capxul in two ways, and neither depends on anyone else.
One command starts a complete private copy: a working copy of the blockchain, the payment services, and the full application. Not a simulation — a fork of the real chain, so every address and contract that exists in the real world exists in the copy.
A real deployed environment we own and deploy to, where the full journey is proven against real infrastructure before anything ships.
This is what makes building with agents fast. Before, everyone shared one test environment. Two people working at once got in each other's way, and one mistake took it down for everybody — which happened to us this month.
They run side by side and never touch each other. No queue, no collisions, no waiting. An agent can try something, break it, throw the whole environment away, and have another one seconds later. It can also move time forward on its private blockchain, so testing something that pays out over a month does not take a month.
Indexing needs rebuilding. That is the part that watches the blockchain for money arriving from outside Capxul and attributes it to the right account. It exists, it works, and it is the wrong shape — so it gets redone.
It does not touch the core. Paying between Capxul accounts, which is the main path, does not need it. It matters when somebody sends money in from an outside wallet.
Two things are genuinely not built yet, and both sit outside the core: the machinery for taking money out to a bank or mobile money, and the machinery for bridging between chains. The routing layer that both of those plug into is built — whether money is going to another chain, a bank, or a different provider, the part that decides how it gets there already exists. What is missing is the connections on the far end.
Standing up a production environment is also still to do. That is provisioning, not building — the code is ready for it.
This is the part worth understanding, because it is what a finished foundation buys you.
The last three weeks were one enormous piece of work where everything depended on everything else. Nothing could be finished until the thing underneath it was finished. That is why it stalled repeatedly, and it is why one bad week cost so much.
That shape is gone. From here the work arrives as small, separate pieces. Each one is a complete thing a person can do — send money, run payroll, send an invoice — and each takes a couple of days rather than weeks. They can run side by side. If one hits a problem, the others keep moving.
Two rules keep it that way. We build against screens that already exist — the designs are built, so we connect a screen that does nothing to something real rather than inventing new surfaces. And nothing starts until its unknowns are named — each piece is mapped first, and the open questions get answered before code is written. Ignoring that second rule is precisely what cost us three weeks.
Ranges rather than dates, for one honest reason: we have not yet completed a capability under this new way of working. The first one is also the measurement. After it lands we can give real dates.
Three weeks of rebuilt platform moves onto the main line. This releases feature work, which is currently held back on purpose.
Two start immediately and neither blocks the other. This is where we learn our real pace.
Payments and history, sending money, invoices, contacts, payroll, and the inbox where an invoice sent to you arrives. Several run at the same time. After that, a person can do everything the core product promises.
One assumption: that no supplier withdraws a product underneath us again. Not hypothetical — it happened once in the last three weeks and cost a day.