To tell it to alter its responses, they would need to recognize that what they expect is incorrect. (Assuming we trust LMMs to tell us the truth, which we shouldn't, but I think that's orthogonal to this.)
I just get alerts over telegram. They're just regular notifications, but you could probably set specific notification behavior and sounds for the app?
Edit: Though that's not a selfhosted solution. Sometimes it's a decision between reliability and selfhosted. e.g., I went back to Tailscale from Headscale when I lost connectivity through Headscale and couldn't figure out how to get it back.
Thank you for posting this! I've been increasing the memory for my VM over and over and it was using 24GB RAM + 4GB swap. Hopefully this will let me reclaim some.
Natto is one of the worst flavors and textures I've ever had in my life.
Roasted natto is not bad.