I launched a paid dev tool with no free trial. Here's the math that made me do it.

Most advice says give people a free tier. I didn't, and I want to lay out the reasoning honestly, including the part where it might be wrong.

The product is a hosted Linux machine that AI coding agents run on, so your work keeps going when you close the laptop. Five tiers, $19 to $199 a month. It started as an open-source project I self-hosted, which is still up and still free.

Here's the thing that killed the free tier before I ever built one.

Every account is a real virtual machine. Not a row in a database, not a seat, not an API call I can rate-limit down to nothing. A VM with CPU and RAM and a disk attached, and it costs me money the minute it exists. A free user and a paying user cost me nearly the same amount to serve.

That's the whole difference between this and a SaaS app. If you sell software, your marginal free user is close to free. If you sell compute, your marginal free user is a bill.

So the options were: 1) charge from day one, 2) run a free tier I quietly subsidize and resent, or 3) build an aggressive idle-reaper that kills free machines fast, which would break the one thing the product is for. Option 3 is the funny one. The entire pitch is that your machine doesn't stop when you walk away, and a free tier would have to stop when you walk away.

So: no free trial, and a 7-day refund instead. The exact wording on my site is "no form, no call, you email me." That's the whole policy and I'd rather it be embarrassingly simple than have a cancellation flow with a retention offer in it.

The other decision people ask about is the AI billing. You bring your own Anthropic or OpenAI key and they bill you directly. I don't mark up tokens and I don't proxy them. Reselling inference would be a decent margin and I'd rather sell one thing well than have a second business I don't want to run.

Every feature is on every tier, too. The tiers only differ by CPU, RAM and disk. Paying more buys horsepower, not permission.

Where I'm honest about the risk: this makes the first click much heavier. Nobody gets to poke at it for ten minutes and decide. On Hacker News today the first thing I did was raise that objection myself, because it's the correct objection, and I don't have a good counter beyond "here's a refund with no friction on it."

Current state, since I'd rather be specific than vague: one paying customer, who signed up today. That's it. So treat all of this as reasoning I can defend rather than a result I can prove.

What I'd genuinely like to know from people further along: if you sell something with real per-user infrastructure cost, what did you do? Did you eat a free tier and grow into it, or did you charge from the start and accept the slower top of funnel?

I picked the second one because I could see the math on the first. But I've only got one data point, and it's a week old.