MetaEnd

Proton Pass shipped AI access tokens: scoped, read-only, audited, revocable. The credential layer of agent security finally has a real answer.

A second reader has arrived and it cannot see your homepage. metaend Grade scans any site for AI agent readiness and grades it A to F.

The architecture and security choices behind an always-on personal AI assistant: threat model, ZDR inference, sandboxing, encrypted brain.

One HTTP endpoint. $0.01 a call. No signup, no API key, no account. Built on x402 for agents that need to fix broken JSON and just want to get on with it. Every LLM pipeline I have ever touched has the same quiet bug. Somewhere, a model hands back JSON that does not parse. A trailing comma. Single quotes instead of double. A response wrapped in a markdown code fence. An object truncated mid-key because the generation hit a token limit. So you write a cleanup function. Then a slightly bigger o...

Markdown wins for instructing LLMs. HTML wins for outputs humans read. A paste-ready prompt for living spec artifacts in any project.

How to prove something about private data without revealing it, when AI and regulation collide. Why we ended up on Aztec.

DOM reflow eats 14.7% of your frame budget during AI streaming. Pretext cuts that to 0.36%. Real benchmarks, 1000 iterations, 5 scenarios.

AI agents can't tell if you're human. A2H Proof flips verification: the agent challenges, the human proves, the math checks out.

GEO has real research behind it, but the hype outpaces the evidence. Here's what actually works, what doesn't, and why most advice is premature.

Your prompts fail for the same reasons requirements docs fail. REprompt applies classical RE to fix them before the agent ever runs.