Price the first month of a tiny AI-built app.

The invoice for a one-night UI is usually the coding agent, not a rack of servers. Add the three lines below before you let a deploy plan talk you into a cloud account.

Line 1. The tool that writes the app

Check the vendor’s pricing page on the day you start. As of 7 October 2026, the public individual prices looked like this:

• Cursor Pro, $20 a month. Pro Plus is $60 and Ultra is $200 on Cursor’s own plan table. Those higher tiers are for people who hit agent limits, not for a page that needs a home.
• Claude Pro, $20 a month, or $17 a month if you pay $200 for the year. Pro includes Claude Code. Max starts at $100 when you want a multiple of Pro usage.

Pick one plan to put in the sheet, the one you will actually send prompts from. A second subscription “just in case” doubles this line for a project that does not have users yet. Free tiers are a real option if you are still learning the tool. Write $0 and also write what you will do when the free limit stops you mid-page.

Seat price is not the whole coding bill. Through 2026, more of these tools have metered the heavy agent work on top of the subscription. If your dashboard shows usage charges, add last month’s overage as its own row. A quiet month on Pro and a month of overnight agents are not the same number.

Line 2. Tokens, if the app itself calls a model

Many vibe-coded pages never call a model after the agent is done. The page is HTML. This line is $0. Write that down so you do not invent an inference budget out of habit.

If the live page does call a model, estimate with a count you can explain:

1. Average prompt size plus average reply, in tokens. The provider’s tokenizer or a rough 4 characters per token is enough for a first pass.
2. How many of those calls a real visitor makes. A demo you click ten times is not “a thousand users.”
3. The published price per million input tokens and per million output tokens for the model you named in the code.

Multiply. Then look at the result next to line 1. For a personal project, line 2 is often smaller than the subscription until the page is public and busy. If line 2 is already larger than you want, the model is in the request path too often. Cache repeated answers, or stop calling a frontier model for a job a short template could do.

Line 3. A place for the page to stay up

This is the line people either ignore or explode. Ignoring it means the app lives on localhost and the month costs whatever line 1 cost, with nothing to send. Exploding it means a virtual machine, a disk, an IP address, and a certificate workflow for a folder of static files.

For a static UI and a small table of form rows, price the smallest host that gives you HTTPS and one address. On 7 October 2026, dayzero.run’s early access plan was $0, with no card: one subdomain, static files up to 5 MB, and 500 MB of traffic in the month, including a live tunnel inside that traffic cap. That is a concrete $0 you can put on line 3 if the page fits. If your folder is larger than a few megabytes, or you already know you need many projects and a custom domain, write the paid tier you would actually move to and do not pretend the free cap is infinite.

Leave GPUs, private networking, and retrieval off this sheet. They are a different product. Adding them here is how a one-month experiment inherits an enterprise diagram.

Add the column that keeps you honest

Recheck the vendor pages when you reuse the sheet. The $20 figures above were public list prices on 7 October 2026, and vendors change them. The shape of the sheet does not change. One tool, the calls the live page makes, and the smallest address that stays up when the laptop does not.