Four walls, and only one of them is code.

Wall one — post-quantum cryptography, in production, verified on your phone

"Harvest now, decrypt later" is not a hypothetical. Encrypted traffic captured today can be stored until a quantum computer can open it. Every serious messenger is moving; the question is how far each has actually got.

The third row is the one to look at. Post-quantum authentication is the leg neither Signal nor Apple has shipped, and both say so themselves. Signal uses classical signatures; Apple's PQ3 uses classical signatures and states it will continue to assess the need. Our safety number commits to both key sets, so a verified contact is verified against the post-quantum key too.

And the way we prove it is unusual. Signal and Apple verify their cryptographic primitives when they build the software. We verify ours on your device, at every launch, against the standards body's own published test vectors — copied verbatim, provenance recorded per file, none invented. A device that cannot prove a primitive publishes nothing post-quantum and accepts nothing post-quantum. It fails closed, in your hand, every time you open the app.

Wall two — anonymous payments, which we believe nobody else has shipped

This is the wall we would defend hardest, because it solves the problem that defeats every other paid privacy product. Threema spent a decade building an app that never learns who you are, and then sells it through an app store, and the purchase re-attaches the identity the app worked so hard to avoid.

Our rail severs it cryptographically. Your device prepares a token and blinds it. Our server signs a value it cannot read. You unblind the result and hold a valid entitlement that our server can verify and cannot trace back to the payment that created it. It is an RSA blind signature, it runs on our own rail through UPI and cards, and a real purchase has completed the full chain.

The commercial consequence is larger than the privacy one. Because the rail is ours, a sale on the web pays a payment gateway around three per cent instead of a store's fifteen to thirty. On a ₹499 subscription a store takes ₹75 to ₹150 per paying user per year — several times what the entire infrastructure costs to serve them. The payment architecture is a bigger margin lever than the server architecture, and we already built it.

Wall three — a cost base that makes the price survivable

Holding nothing is not a tax we pay for our values. It is a structurally cheaper business that happens also to be the right thing.

• No storage bill. Almost every messenger's infrastructure cost is dominated by retained media. We delete on delivery, so we pay for transit and not for keeping.
• No breach-notification exposure, no retention bill, no analytics pipeline, no ad-tech stack. Every one of those is a department somewhere else.
• No account-recovery support function. With no accounts and no ability to recover one, we cannot be drawn into the most expensive support operation every consumer app funds. Nobody names this and it is worth real money — and the honest counterweight is that the cost we avoid is paid by the user in lost identities unless recovery genuinely works, which is why our backup and restore flows are an economic item and not only a design one.
• Our own relay for calls, not a third-party platform. Relayed calls are the one line that scales with usage and can genuinely bite, and putting calls behind the paid tier aligns that cost with the revenue that pays for it.

Wall four — the record of telling the truth about ourselves

This is the least tangible wall and, over ten years, probably the strongest. Trust is the product, and it compounds only if it is never spent.

Some of it is process. Our build pipeline carries a gate that refuses to publish our website if it makes a tier claim no code gate enforces — it caught us selling two features we then declined to deliver, and it is why that cannot happen again. Our competitor analysis carries a footnote we wrote ourselves saying that two rows overstated us and that Signal beats us on the first one. Our model page says which numbers are measurements and which are our opinion.

The rest is the plan. An independent audit, then the source opened, then the audit report published, then a bug bounty on code that has actually been reviewed. Each turn makes the next one cheaper and the brand harder to copy. Our own note on this is blunt: an AI can reproduce our functions; it cannot reproduce the fact that we shipped them, audited them, and answered for them in public over time.