The filtering part is what I'm most curious about. There's a reddit-like app called Mirage that has an AI feature that checks if a post is nasty and modifies it to be more calm and less offensive lol. It might sound silly but it made my time on that app way better. And using a similar technique you could filter out posts and even customize it by changing the system prompt. I can see using a tiny LLM (perhaps one from the LFM2.5 line, they're super efficient) that someone could run. It could even run in-browser or on-device, probably.