Oh yeah, I’ve got a box of esp32s and mmwave sensors waiting for me to get my soldering iron out. I’m with you, they’re great. I suppose I could somehow figure out out how to do SDR with an ESP32, but I’ve got plenty of unfinished projects waiting for me as it is. ESPHOME makes edge devices so easy to create. And it gives me an excuse to do a little soldering and 3d printing.

The ESP32s are also really good for making old devices smart. One project in my queue is turning a bunch of air purifiers into smart devices that HA controls. They were super cheap and there’s nothing to them but a fan, and a filter but they’re loud. I’m going to make it so they only run when no one is in the room.

I was thinking the same thing. My path was self hosting on old desktops for the longest time then the pi came out and I ended up doing a birdnet project, then home assistant and now I’ve got about a half dozen data collectors on pi zero 2 w machines. They are pretty much set and forget. I use an orange pi for my instance of Solar Assistant because I wanted something that was not affected by anything but the network and the inverter and I know how rock solid these little machines are.

But in the last few years I’ve started embracing docker and NUCs and have pulled apps off Synology, making it just a file server at this point.

So technically I have more raspberry pis than anything else but they are really just edge devices, except for Home Assistant. And it will probably stay that way until there is a good reason to switch. I don’t see that happening for years.

AI Pioneer 🤖

• Fleet: 6–10 devices · 21–50 TB
• Motivation: Privacy & data ownership, Independence from Big Tech & subscriptions, Learning & experimenting
• My MVPs: Portainer, Open WebUI, Ollama, Nginx Proxy Manager, arr stack

deleted by creator

I’ve been running birdnetpi on a raspberry pi 3 for years. It runs very well on this modest hardware. I also run birdnet-go in a docker container and use my reolink security cameras for the RSTP audio feed to that instance.

There are all sorts of AI/ML models available that can run on smaller hardware. But they are usually very specific and not general purpose, like the LLMs.

I hope I didn’t come off as complaining about degoog. I’m definitely not. degoog is awesome.

Thanks for the advice about the proxies. I’ll be looking into that as I continue to work on keeping prying eyes off our internet usage. I suspect this will be a never ending struggle for the rest of my life.

It seems that both google and reddit block requests coming from my VPN provider(PIA). I may have to set up my own VPN server using Azure or AWS if things keep going this way.

Thanks for this! I included degoog in my arr stack and it works great! I was surprised by the language support when I pointed my gluetun container to an an Italian exit node. But it’s very well documented and the language environment variable was easy to set. Nice!

Getting this working within minutes of reading this post felt like opening an unexpected present. I love it.

Thanks for this. I’m definitely dropping ollama now. No wonder GGUF models always gave me issues.

I might even dump open webui for llama.cpp’s webui