𝗖𝗬𝗥𝗨𝗦

Ngl, I expected zkTLS to be one of those things that sounds sick on CT and is painful to use. I've been on Primus since , I've done the tasks, and it wasn't that. Then I went down the FHE rabbit hole and it got more interesting. Start with zkTLS, the part I actually used. I didn't have to ask any platform for anything. No integration, no partnership. zkTLS works with the websites, APIs and exchanges that already exist, and most privacy tech dies at exactly that step. So I ran my own flow. The extension generated the proof from my own browser session. I never shared a login and never sent a screenshot. The app learned one fact, and the rest of my data stayed with me. Every other time I've had to "prove" something in crypto, it meant screenshots anyone can edit or access I shouldn't hand over. A proof is neither fakeable nor risky. For builders, there are two modes. MPC gives stronger protection during TLS session setup. Proxy is faster, with different trust assumptions. You pick the trade-off based on what you're proving. That's the verify half. Here's the compute half. zkTLS proves a fact is real. FHE lets you use it while it stays encrypted. It never took off because it was too slow, and Primus published a technical deep dive on how they fixed that: → A full confidential transfer on encrypted 64-bit balances finishes in about 90 ms → One bootstrapping takes about 5 ms → 210+ confidential transfers per second on 8 consumer RTX 5090 GPUs, scaling linearly The problem: classic FHE processes one tiny gate at a time, and every gate needs a noise refresh called bootstrapping. One 64-bit operation becomes thousands of them. The fix: pack more work into each bootstrapping, then use AutoHoG, a compiler that takes standard Verilog and collapses clusters of gates into single custom gates, cutting latency by up to 5.7x. The GPU engine then runs one uniform operation, which is exactly what GPUs are good at. One detail I thought was underrated: they use exact integer arithmetic instead of floating-point, so results are byte-identical across machines. For decentralized consensus, that matters a lot. And it runs on consumer GPUs you can buy off the shelf. Privacy usually loses because it costs too much. When a confidential transfer takes about 90 milliseconds, privacy stops being a feature you have to justify and becomes the default. Put the two halves together: zkTLS proves facts from the platforms people already use, and FHE computes on the result without decrypting it. Both are getting fast enough to feel like normal software. With 1M+ users and 10+ paid enterprise customers, this isn't staying in the lab. Proof without exposure isn't a slogan once you've used it yourself. What would it take for you to use privacy by default: speed, cost, or not having to think about it? @primus_labs