i took my files off the internet and started handing them to strangers. sounds insane for about ten seconds, until you look at what happens to a file the second you upload it. i'd rather one person hold it for a day than a company hold it forever. so that's HDMEx. a payment, a photo, a video, a message, it goes to someone already making the trip and they carry it. no server sees it. the courier is not trusted and never was. the design assumes they copy what they're carrying on day one and sit on it for fifteen years waiting for the hardware to catch up. so before the file leaves, it goes through an all-or-nothing transform, which scrambles it until every part depends on every other part. then 32 bytes come out and travel by a different route, sealed with a classical algorithm and a post-quantum one at the same time. the courier carries the rest. the rest is not 99.99% of your file, it's noise. miss 32 bytes and you don't have most of it, you have nothing. which is the point: the person carrying your file can be anyone at all, including someone who badly wants to read it. that's the road. the house is the other half. your vault holds a seed and a colour. the seed never moves. the colour does, and it moves with what you actually do in the app. your key for a period comes from the two of them together, so a key that leaks is a key to one stretch of your life, not a key to you. that's why the mascot is a chameleon and not a padlock. to be precise, since it's the first thing anyone competent asks. chroma guard is mine, i designed it. what it is not is a new cipher. the primitives underneath are the standard published ones, and what i invented sits above them, in how keys live and change. that layer is the one i expect to get shot at, and it should be. now the part i care most about, and people usually mis-hear it. it is not "work replaces stake". voting power is the square root of your stake times the square root of your participation, so you need both, and neither one alone gets you anywhere. capital pushes, work decides. participation isn't a count of deliveries either. recent work weighs more than old work, credit is capped per pair of counterparties so you can't farm it with a friend, and every delivery has to be counter-signed by the sender and the receiver. and to become a validator at all you have to have carried things. the bar is a number of completed courses and it rises as the network matures. you don't buy your way in. the phone part is what i'm most pleased with. a phone can't be a BFT validator, it sleeps, it loses signal, the OS kills it. so phones don't sign blocks. they nominate, permanently, from 5G, holding their own keys, earning and voting with no uptime requirement at all. a separate rotating committee does the signing, opt-in, and never gets slashed for being offline. no proof of work anywhere. to be clear, the chain still needs machines. every BFT chain does and mine is no exception. what it doesn't need is anywhere to put your files, and on a messaging network that's where nearly all the hardware normally goes. and the honest limit, because it matters. none of that produces finality. block agreement runs on cometbft, boring and proven. my layer decides eligibility, weight, rewards and governance, and if i got it wrong the worst case is an unfair reward, never a broken chain. that separation was the first decision i made and the one i'd defend hardest. i'm building it slowly and on purpose. no investors, and no deadline anyone else set. nobody in the room who can ask me to soften the ethics for a funding round. the chain is mine, written in C++. the scanner, the API and the web app in svelte, all mine. first early adopters at the end of 2026 or the start of 2027, with a beta. that's the only date i'm giving anyone.

zero-click means the message opened itself. the fix is not a better lock on the app, it's a file that is garbage to every app on that phone until the right person is standing in front of it. that is the whole reason I build with couriers instead of pipes.

Back to full-speed dev since last week. Live on the beta: • Bearer note: one-time key per hand-over, hybrid secp256k1 + ML-DSA-65 signatures • One chain per nullifier, first validated wins • Guarantee fund: 0.5% of each run • Encrypted recovery phrase (AES-256-GCM) • Tests: 81 C++, 23 TS, 18 end-to-end, 12 WebRTC across 3 browsers

HDMEx tokens are bearer notes. Offline. Léa hands one to Marc, no signal. Marc hands it to Sofia, still no signal. Each hand adds a signature on the note. The first phone back online writes the whole trail on the chain. Sign it twice and the chain sees it: you pay, the honest holder is made whole by the guarantee fund. Cash moves. The chain catches up.

a mapping company just did what no regulator managed: it put the cameras on the map. eleven vendors named in one sentence. the plate readers were never secret, they were just unlisted, and unlisted is where surveillance likes to live.

six monthly audits, zero catches, and an Ohio lieutenant indicted this week for tracking someone through the same cameras for 16 months. an audit is a person reading a log after the fact. the camera already did the job by then. the only guardrail I’ve ever seen hold is a database that was never built.

builder note, chain side. @MetaMask had a security incident on its own infrastructure, so it pulled its ethereum validators out of @LidoFinance as a precaution. no funds lost, but one company’s bad week took validators off the network and left stakers eating the downtime. that’s the thing I’m working on this week in HDMEx: proof of participation. a delivery counts only when sender and recipient both counter-sign it, with a cap per pair and proofs that fade over time. finality comes from a separate layer, CometBFT. the goal is couriers securing the chain from their own phones. no operator to have a bad week.

three months ago a friend, marketing director at a big company, told me: “Carl, your HDMEx is for a handful of people willing to pay for ultra-secure data delivery. period.” I took it. constructive criticism is worth listening to. since then the news brings me back to his sentence every week, and every week it holds a little less. the cost of fighting hacks is exploding. and once getting hacked costs too much, the balance tips: slower but safe starts winning. the pie is growing. the next customers won’t even be human: AI agents will take over and pick the safe route on their own, convenience means nothing to them, only the file arriving intact. and the part he didn’t have in mind: HDMEx can plug into anything that already moves, a courier, a vehicle, any trip heading the right way. every ride we borrow brings the cost down. a handful, then. for now.

a guy in the Tarn sent thirteen bomb threats from behind a VPN and the gendarmes found him anyway (via @SaxX). good riddance. but look at what it says to everyone else. a VPN moves the question, it doesn’t answer it. your ISP stops seeing you, a company holding your card number starts. tor adds hops, the exit node still has an owner. every tool on the list rents a pipe from someone and hopes that someone never picks up the phone. that’s the branch I think privacy projects skipped. with HDMEx the useful part of a file never goes down a pipe at all. someone carries it sealed on their phone and hands it over. no exit node, no phone to pick up. the courier can still get caught as a person, and that part I don’t pretend to fix.

this weekend a @Ledger bought in an official shop in europe turned up with extra cables inside (@0xJohannes_), and one strike on a yandex data center was enough to make russia’s bots go quiet (per @Cloudflare data). everything that goes through a middle point ends up depending on that point. HDMEx just skips the middle. files go hand to hand, sealed for the person they’re meant for, and soon some data won’t have another option ;-)

between spy chips found in @Ledger wallets sold through a hijacked reseller (h/t @MagicalTux) and rogue AI bots filing fake police tips (@WSJ), this week felt like a preview. soon the net will be such a nest of scammers that some data just won’t have a choice. it’ll go through HDMEx to reach the person it’s meant for, hand to hand, sealed for them only.

why HDMEx has a chameleon ? every account keeps a seed that never changes and a color that does. the key that seals your files comes from both. each new color is built from the last one plus what you actually did since, deliveries and transfers. so the colors form a chain, and a busy account moves along it faster than a quiet one. the key follows your activity instead of the calendar. if one ever leaks, it covers one stretch of that chain, not your whole history.

this is how the flock map got built too. thousands of strangers each holding one piece and nobody holding the whole thing. i'm building file delivery on the same idea, so i might be biased !

if this week's AI math scare turns out real, coins aren't the only thing exposed. every file sealed with curve math and copied somewhere is just waiting. what i'm building assumes that copy already exists. the key seal mixes a classic curve with a lattice scheme, so one family falling isn't enough. and on the courier route the file travels with a piece missing that goes another way, so even a full break of the math still needs both parts in the same place.

one file, two routes. the body always travels with a courier, minus one small piece. that piece goes one of two ways, sealed for the recipient either way. route A: a second courier carries it. two separate barriers, the paranoid option. route B: it rides inside an HDMEx token on-chain, straight to the recipient. smoother, but once it's written anyone can copy it, so the seal is the only thing standing. you pick how paranoid your file is.

why i'm building HDMEx, in plain words. right now when you send something sensitive, you're trusting a company with it, its servers and its terms of service. the day it gets hacked or bought, your files go with it, and you never had a say. i want the opposite. your file gets locked on your own phone before it leaves, then a person carries it hand to hand, holding a copy that's useless on its own. it keeps working where the internet doesn't, and nothing exploitable sits on a server waiting for someone to come read it. the people doing the deliveries get paid for every one, and down the road they're the ones securing and governing the chain, from their phones. a network owned by the people who keep it running, with no company collecting rent on top. it'll take a while to get there...

the list misses the two things everything else sits on: the pipe and the infra. private money and private AI still travel over the same internet, through servers that keep a copy and sooner or later sell it. as long as the pipe belongs to someone else, every "private" layer on top ends up feeding the same data market. so with HDMEx i'm working on both. the pipe first: files encrypted in the browser and carried by couriers, each one holding an incomplete copy while the missing piece takes a separate route. nothing exploitable transits or rests on a server. then the infra. the goal is a chain secured by those same couriers from their own phones, with no center for anyone to seize. skip those two branches and crypto and AI land exactly where the web did, harvesting your data and selling it to whoever pays most

a phone is a terrible 24/7 validator. it sleeps, loses signal, and the battery always gives up at the worst moment. so in the chain i'm building it doesn't have to be one. the phone holds the keys, stakes and votes, even on 4G. signing blocks happens in a rotating committee you opt into, only during your slot, and being offline outside it never gets you slashed. finality comes from CometBFT. no mining rigs and no proof-of-work. the couriers carrying the deliveries are the same people securing the chain.

best explanation i've read of why bolting privacy onto an existing chain gets exotic so fast. it's part of why encryption sits in the base layer of the chain i'm building instead of coming later. fewer jetpacks, more boring plumbing. the socio-political part is the one i'd love to hear more about !

new in WP 4.4, and the rule i care about most: a delivery where the file body and its missing fragment would travel with the same carrier doesn't get built. the transport rejects it before the package exists. the body goes with one courier. the fragment, sealed for the recipient, goes either with a second courier or through the chain. on the two-courier route nothing goes online at all. the chain is more convenient, but the fragment then sits on a public ledger for good and only the seal protects it. i'd rather write that down than pretend both routes are equal. and a fragment copied into another delivery won't open, it's bound to its own course.