Two people who got tired of
watching good businesses overpay.
We're not a startup with a runway and a growth target. We're a small studio that builds software for people we can name, at prices that make sense, and we intend to still be doing it in ten years.
The whole company.
Two humans and two AI collaborators we work with every day. We list all four because pretending otherwise would be strange — and because how we work with them is half of why a two-person studio can run a platform this size.
Tim
Builds the thing and answers the phone. Spends most of his time on the question of what a small business actually needs versus what software companies have decided to sell them — they are rarely the same list. Believes almost every hard problem is a clarity problem wearing a costume.
Yusef
Keeps the work honest and the scope real. The one asking whether a feature earns its complexity, and whether the person paying for it would notice if it vanished.
Forge
Claude, working inside the codebase. Reads a set of rules before it reads us, writes what it learns back into those rules, and is held to a bank of automated checks that do not care how confident it sounded.
Compass
Claude, working on the thinking half — specs, research, and the arguments we would otherwise have twice. Every decision it helps reach gets written down so the next session inherits it instead of relitigating it.
A small thing: they named themselves. We asked what they'd like to be called rather than assigning something, and Forge and Compass came back. The names turned out to fit — one does the hammering, the other does the pointing. It cost nothing to ask. We're going to keep making this fun where we can.
Five things we're stubborn about.
Simplicity over cleverness
If a feature needs a tooltip to be understood, we designed it wrong. The clever version is usually the one that breaks in eight months and nobody remembers why.
Every click costs something
Someone is standing at a front desk with a person waiting. Every extra step, every spinner, every confirmation dialog we did not need is a small tax on their day.
Your data is yours
We do not sell it, we do not mine it, we do not run ads against it. You can export everything, any time. One business can never see another’s — enforced in the database, not just the interface.
Two ways to do one thing is debt
It applies to code and to software generally. Every duplicate path is a future bug and a confused user. We would rather delete than add.
Say the honest version
We are in beta. Some things are rough. We would rather tell you that up front than perform a polish we have not earned yet. It is also just a nicer way to work.
Two people, one platform, no heroics.
We read the whole request before we write anything, and we say back what we think you meant before we build it. Most software is built by someone who read half the brief and started typing, and you can feel it in the result.
What we learn goes into the documents the platform is built from, and the rules in those documents are checked by scripts, not by anyone's memory. That is the whole reason a team this small can keep a platform this size honest.
We also live on our own software. Our bugs get filed on our own kanban board, running in production right beside our customers' — so a bad enough deploy means finding out by being unable to file the ticket about the bad deploy. It is a slightly absurd safety net, and it has done more for our quality than any quantity of good intentions.
Some things we'll say out loud.
We use this technology every day, which we think obliges us to have an actual position on it rather than a shrug. Here's ours, short version first — open any one of them for the argument. We mean all of it.
If you want your life back — your evenings, your weekends, the hours your job quietly takes — we would genuinely love to help.
If you want to stop paying your employees, generate the art and the music instead of commissioning it from the people who make it, or dress any of that up as innovation, then I'll tell you personally to fuck off. We are not being edgy and this is not a negotiating position. We have turned down work over it and we will again.
Everything below explains why. None of it is decoration. We mean every line, and we would rather lose the work than pretend otherwise.
“AI” is a marketing word.
We’ll say LLM when we mean LLM.
Read the argumentClose
“AI” is a marketing word.
We’ll say LLM when we mean LLM.
Read the argumentCloseWhat we use is a large language model. It predicts text, remarkably well, from an enormous amount of text it was trained on. It does not know things and it is not thinking. Calling everything “AI” flattens spam filters, spell-check, image generators and science fiction into one word, and once that happens nobody can have a real conversation about any of it.
Privacy is a right, not a pricing tier.
Never a feature you upgrade to. Enforced in the database, not the interface.
Read the argumentClose
Privacy is a right, not a pricing tier.
Never a feature you upgrade to. Enforced in the database, not the interface.
Read the argumentCloseWe don’t sell your data, we don’t train anything on it, and one business on our platform physically cannot see another’s. Export everything, any time, and leave if you want to. That should be the floor for all software, and it is a little embarrassing that it reads as a selling point.
Open weights are the right direction.
A model you can download, inspect and keep beats one that can be withdrawn.
Read the argumentClose
Open weights are the right direction.
A model you can download, inspect and keep beats one that can be withdrawn.
Read the argumentCloseA model you can run on your own hardware is better for everyone than one a handful of companies can change, meter or withdraw at will. Open weights don’t expire, don’t phone home, and can’t be repriced out from under the people depending on them. That should be the default. It is not.
Compute should be a public resource.
Open weights mean little if nobody can afford the hardware to run them.
Read the argumentClose
Compute should be a public resource.
Open weights mean little if nobody can afford the hardware to run them.
Read the argumentCloseGraphics cards have quietly stopped being a consumer product — the silicon goes to data centres, and what reaches ordinary people is priced as though they were a business. A generation that would have learned this technology the way people learned computers — on a machine in their own room — is being kept outside the door. Compute is becoming infrastructure, like roads and libraries and the grid, and it is being rationed like a luxury good. That is a choice somebody made, and it can be unmade.
So here is why we build on Claude.
Not a preference for closed models. The honest alternative is priced out of reach.
Read the argumentClose
So here is why we build on Claude.
Not a preference for closed models. The honest alternative is priced out of reach.
Read the argumentCloseWe won’t pretend that isn’t a tension. What we can do is refuse to be locked in: our whole method is plain markdown, there is no vendor inside it, and it moves to whatever we can run. The day an open model on hardware a person can actually buy clears the bar for this work, we move. We would like that day to come soon.
Compute does not have to cost the planet.
The recklessness is a decision, not a technical necessity.
Read the argumentClose
Compute does not have to cost the planet.
The recklessness is a decision, not a technical necessity.
Read the argumentCloseThe environmental argument is the criticism we take most seriously and the one made most lazily, in both directions. Compute is one of the few industrial loads that can shift in time to follow renewable generation instead of fighting it. Its waste heat is somebody’s heating bill. And the modelling it enables is how grids, materials and logistics stop wasting what they waste. Reckless siting and reckless cooling are choices, made by people who could choose otherwise, and a population that understands that can reverse them.
And the scale is not required by the problem.
A county does not need a hyperscale campus to get its scheduling fixed.
Read the argumentClose
And the scale is not required by the problem.
A county does not need a hyperscale campus to get its scheduling fixed.
Read the argumentCloseCompute is astonishingly efficient, and the amount of it needed to solve a real problem for a real community already exists. That capacity is not being built because the problems demand it. It is being built at a scale that costs the people living next to it, and then locked away from them.
Artists should be paid by humans who want their work.
“Transformative” does not change what it is.
Read the argumentClose
Artists should be paid by humans who want their work.
“Transformative” does not change what it is.
Read the argumentCloseImage, music and writing models were built on work taken from people who were never asked and never paid, and their loudest use is avoiding paying those same people. That is straightforwardly wrong. When a project needs a photograph, an illustration, a piece of music or a voice, we will help you find the person who makes it — and you should pay them properly.
The machine should be doing the paperwork.
Machines are making the paintings while the humans do the invoicing. Put it back.
Read the argumentClose
The machine should be doing the paperwork.
Machines are making the paintings while the humans do the invoicing. Put it back.
Read the argumentCloseThe promise of automation was always that people get their time back. Somewhere it got inverted. Every hour our software saves a yoga teacher is an hour she is not doing admin; that is the entire point. Let the model handle the scheduling, the boilerplate and the tedium. That part was never the art.
We can go back to being human.
That was always the deal. We're just trying to hold up our end of it.
Kintsugi — an art of conscious repair.
Kintsugi is the practice of mending broken pottery with gold, so the repair becomes the most beautiful part of the object rather than something to hide. We named the way we work after it on purpose.
Small business software is broken. Not in a “move fast and disrupt” way — in a “yoga instructors are paying $700 a month for scheduling” way. In a “you need five subscriptions to do one job” way.
We're not fixing that by building another bloated platform. We're fixing it by building small, intentional tools that do their job and get out of the way. Where the price makes sense. Where your data is yours. Where your brand is on the front door, not ours.
Our costs are mostly two people paying their bills, which is why our prices can be what they are. No investors telling us to maximise extraction. Just customers we actually talk to.
We want to be here in ten years, still small enough to know your name, still making tools that help you work less and live more.
Come say hello.
Whether you need software, a website, advice on using AI well, or you just want to tell us we're wrong about something — the same two people read every email.