Tallyback · Protocol and receipt checker
Your agent says it's done. Tallyback checks.
The record of what AI agents actually did.
How it works
An agent makes a claim, like "I authorized the repair order" or "I recorded the payment." Tallyback runs the same loop every time: observe, act, read back, compare. Then it writes a receipt that says whether the claim held up.
It works in any industry. A small core handles the rules, a profile describes one industry's vocabulary and what counts as proof, and an adapter talks to the system itself, whether that's a shop system, a payment system or a support desk.
I started Tallyback because agents are good at saying they finished something. I wanted a plain way to know when they actually did.
Tallyback is a private prototype, not a production service yet.