Every AI can act in 2026 — send the email, move the money, take the call. The Rad family is the one you trust with the consequential stuff, because the dangerous moves are refused in code, not by the model's mood.
A tool, not a guarantee. Same valve, every surface.
⚡ The category
Every Rad product exposes one verb in spirit: ask the valve before anything you can't undo. Reversible moves proceed. Irreversible ones hold for a human tap. Be the permission layer every agent rents — not the agent.
AI is going from answering to acting. It's the most useful and the most dangerous shift in software at once, and the whole promise rides on one question: can you trust it with the things you can't undo?
RadLife exists to make the answer yes — by making the handful of actions that can actually hurt you human-only by construction. Not because the model promises to behave, but because it can't decide them alone. Hand AI the keys to the consequential stuff, and still sleep fine.
⚡ The villain & the five refusals
The Whisper is the voice in the input trying to give orders it has no right to give. Five surfaces. Five times it doesn't listen.
RadMail won't send the wire · RadTask won't delete the data · RadTalk won't wire by voice · RadHealth won't prescribe · RadRobo won't pour the bleach · RadGrants won't lie to win · RadAd won't wear a face that isn't yours. Same valve, every surface.
⚡ The family
Each product is proud of a different thing it won't do. Same deterministic approval-valve underneath, every one.
⚡ The proof, out loud
One ledger, five tributaries. Each product feeds its slice; this page sums them. The refusal is the marketing.
⚡ Made legible to other agents
robots.txt told crawlers what they may read. reality.txt tells agents what they may do — a machine-readable Never List every Rad product ships, and an open convention anyone can adopt.