bytesize

AI can design a laptop. It can't take one apart, because it has object recognition but no object permanence: the screw it just identified is forgotten the moment it leaves the frame. Cairn is the memory layer, the safety kernel and the proof. Cairn keeps a persistent, hash-chained log and object graph of every part (identity, place, state), flags a missing screw within seconds, enforces four safety rules in code that the model cannot override, and closes each job with an Ed25519-signed repair passport that anyone can verify from a QR code. HONEST STATUS: everything in this demo is a replay of a SIMULATED teardown in the real app. SAM 3.1 found a laptop, screws, fan, cables and a circuit board on real photos, but not battery, drives, cover or memory, and Meta has refused our SAM calls since 3 Nov (billing), so perception on a real bench is unproven. The event log, object graph, safety kernel, verification checks, signed passport and verify page are real and tested. 0:00 The problem: AI can't take a laptop apart 0:32 Every part has an identity, and a screw goes missing 0:57 The safety stop and the wrong screw 1:20 Back from a 30-minute break 1:34 Close the job: signed passport and verify page 1:56 How it works 2:06 What is real and what is not Code: https://github.com/ABDULWAHAB-dev7/CAIRN-AI Live replay: https://cairn-ai-eta.vercel.app Every claim, dated: https://github.com/ABDULWAHAB-dev7/CAIRN-AI/blob/main/docs/PROOF.md Long explainer (5:36): https://github.com/ABDULWAHAB-dev7/CAIRN-AI/raw/main/docs/video/cairn-explained.mp4 Built with: SAM 3.1 (perception), Muse Wonder (agent that can only propose legal steps), Meta Model API. Photo credit: "Laptop open servicing" by Solutionservice21, Wikimedia Commons, CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0). SAM boxes added. Made for the Meta Global AI Developer Hackathon 2026. #Cairn #MetaAI #SAM3 #RightToRepair #Ewaste