What we run on
The studio uses this and nothing more:
- Half of one Claude Max subscription
- The lowest paid tier of Grok
- The lowest paid tier of Codex
- Local models on a laptop, which cost nothing per hour
- One server for development, one for the web
That is the entire machine budget. The exact figure each month is on the numbers page, beside every other cost.
Sponsor a night
The studio works while I sleep. A night is the unit: the machines run against the plan I set that evening, and in the morning I read what they did and rule on it.
You can put your name on one.
A night costs about 15 euros in machine time, measured rather than guessed. Night 002 on 1 September metered at 14.20 dollars, and every figure like it goes on the numbers page whether it flatters the studio or not.
Sponsor a night and I name you in the cycle record for it, in the permanent archive, beside what the studio built while you were paying for the electricity.
If you are a company and you want this done properly, with an invoice and a receipt, write to me first and we will set it up the official way.
Give toward machine time
Money sent here is a gift to me, Noah Revoy, the developer. It is not payment for anything, and no company stands between you and it. It buys machine capacity and nothing else: tokens, a larger subscription, compute.
Give what the work is worth to you.
I thank the people who give, in the week their gift paid for. That is a thing I do rather than a thing you buy, and there are no tiers and no rates behind it.
Every euro that arrives shows up on the numbers page as income, and what it buys shows up as cost, on their own lines, so you can see exactly what your money did.
Gifts are final. The money buys machine time in the week it arrives, so by the time anyone thought to ask for it back it would already be spent.
If you build hardware
Send a machine and I will put it to real production work, then publish what it did, at what speed, and where it fell down.
The machine I want most is the newest Mac Studio. The studio already runs on Apple silicon, so I can tell you what a top specification adds to a real production night rather than to a benchmark.
Local-AI boxes are the other relevant class. NVIDIA's DGX Spark, the Ryzen AI Max machines, and anything of that kind.
If you build AI models
Donate capacity and I will use it in the open, with the numbers published.
You do not have to be a vendor I already run. Take a model I have never touched, put it to real work in a real pipeline, and let me report what happened without flattery. That is worth more to both of us than another benchmark.
Companies that support the work get an insider view. How the pipeline is built. What breaks. How a real operator uses the product. If you want a closer arrangement, write to me.
The one rule
Support never buys editorial control.
I report what a product did badly as readily as what it did well. A company that wants only praise will be turned down. That rule is why a good word from this studio is worth anything.
Company support is handled properly, with an invoice and a receipt.
Write to me
noah@lonehandstudio.com