Contrast with https://frigate.video/, which is a locally installed NVR server that pulls camera feeds over the LAN (from a very wide range of off-the-shelf IP cameras) and does all kinds of really neat local processing to do things like (optionally hardware-accelerated) object and audio detection, face recognition, ALPR, semantic search over recorded video, and more — while still maintaining similar privacy guarantees.

It's great that you've done reproducible builds for camera firmware, since that means you don't have to trust a shady IP camera vendor to be competent. Of course, with off-the-shelf stuff, you can largely avoid the security issues there by putting your cameras on a VLAN that can only reach your NVR.

What I don't get is why there needs to be a cloud relay involved at all. If you're fully E2E encrypted anyway, just have the app communicate directly with the camera via STUN.

I see you're planning on selling the preassembled hardware. There's definitely something to be said for "buy this device, download app, done" ease of setup for the wider market that meaningfully improves their privacy over Ring/Nest/et al. But for the power user and self-hosting crowd, I think Frigate makes a lot more sense.

Do you think it would be possible to use ESP32 (RISC-V CPUs) based cameras?

Both for cost reduction and availability of the hardware reasons.

Maybe with a ChaCha20-based cipher instead of AES?

I still think thingino needs a lot, but it does the job so far.