Joel - Nostr Hypermedia

Trusting AI companies with our data is a mistake. Running open-weight models locally is one of the most powerful ways we have to reclaim our privacy. The problem is that most of us can't run trillion-parameter models locally without expensive hardware. So I've been looking for ways to access these models without sacrificing privacy, and found that Trusted Execution Environments (TEE) were the right fit for me. For those unfamiliar: TEE means your data is encrypted inside the processor itself while the model runs. The provider can't see it, can't access it, and can't modify it. Trust moves from their promise to the hardware. I found @PayPerQ , where I can pay with Monero and use their proxy for their private/* models (including Kimi K3, GLM 5.2, DeepSeek v4 Flash, among others) which run inside TEEs, without even creating an account. Of course, I wasn't just going to take their word for it so I ran a security audit using Kimi K3 on the private mode proxy (https://github.com/PayPerQ/ppq-private-mode-proxy), and the tinfoil.sh SDK (their provider for private AI models) to make sure it's legit. The audit surfaced 10 findings, which I've already reported to Matt (founder of PayPerQ), but it looks pretty good. Curious to hear if anyone else is experimenting with TEE AI. #nostr #ai #privacy #monero #security