It was supposed to be "just fix the menu". Along the way you changed the navigation, reverted the navigation and asked for it to be done properly after all. The agent announced several times that it was perfect now.
For an evening like that you can get a receipt. With the number of prompts, token usage and a total at the bottom. Among the line items you'll even find how many times the agent wrote "You're absolutely right". Apparently agreeing with you was part of the job too.
It's made by VibeReceipt, a small independent project for people using Claude Code, Codex CLI or OpenCode. It collects information from the history stored on your computer and lays it out like a printed store receipt.
The total shows what the recorded usage would cost at API prices. If you're on a subscription, you don't get an extra bill. Regular chats with Claude in the browser aren't covered by the tool.
What actually ends up on the receipt
Sessions, your prompts, agent responses, tools used and the number of files edited. Plus tokens, the chunks of content the model processes, and a breakdown of usage and cost per model.
The most expensive model plays the cashier. There's a receipt number, a date and a barcode. You get the whole thing as a PNG, so you can post it to a group chat without screenshotting your terminal.
The most fun items are the less accounting-like ones: how many times you interrupted the agent, how often it replied "Perfect!", what time of day you work with it the most. There's also a frustration counter that looks for selected English swear words in your prompts. Swear in another language and it won't notice, so you can come across as remarkably calm, even if your chat history suggests otherwise.
No account needed and nothing gets uploaded. The analysis runs locally, and the content of your conversations isn't printed on the image. You can add project names separately if you want to show them.
A big total doesn't mean that's what you paid
On a subscription it's easy to stop thinking about how much work a single "try again" kicks off. VibeReceipt puts a number on it. It takes the recorded usage and converts it at API rates, accounting for different token categories, including reused context.
It can also compare the result with the price of your plan. Just don't announce right away that you saved a few thousand. If the summary covers several services or some separately paid work, one subscription won't describe all your spending.
In the same way, a big receipt says little about the value of the app. The agent might have built a working feature or spent an hour fixing its own fix. Usage went up either way.
And that's exactly why this idea is more interesting than yet another token counter buried in the settings. You get a short summary of your work with AI, weird habits included. People already show off their app, stack and revenue in group chats. Now there's a receipt for getting to the "really done this time" version.