Skip to Main Content

AgentTab keeps tabs on what an AI agent spends, and on the rules it spends by. It is a harness for an agent that buys small metered resources while nobody is watching. The agent follows a versioned rule (v1: pay per article). When the seller’s offer changes, the harness notices, proposes a new rule with the numbers, and asks a person before switching, by voice or one click. It cannot raise its own ceilings: a proposal that loosens a hard limit is rejected unasked. If a new rule turns out to cost more per call than the last one, the harness rolls itself back. Payments are real transactions on the Tempo testnet over the Machine Payments Protocol (HTTP 402); the articles are made up. Problem it solves Agents are starting to spend money on tools, data and APIs unattended, and two things go wrong. Terms drift (a seller adds a tab or changes a price), so a fixed config goes stale, and a spending limit written in a prompt is only a suggestion. AgentTab moves the rules outside the model: they are versioned, changes need approval, the agent can’t loosen hard ceilings, and it undoes its own changes when its measured cost per call says they were wrong. Payment decisions are deterministic code, not an LLM. Demo: https://sandboxlabs.ai/agent-tab/ Explainer: https://sandboxlabs.ai/agent-tab-explainer/

Team

More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon

MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

MongoMation project preview

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

LabRat project preview

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Recursive Harness Scientist project preview

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

SuperBid project preview

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…

WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…

WorldSTATE project preview