I feel a bit stuck with my private LLM projects lately.
I’ve built some useful things: a daily news briefing from my feeds and an inbox cleanup that filters spam-like emails. It works, but it’s not what I actually had in mind when I started.
What I really want is something more personal. A system that knows me. Something that greets me in the morning, looks at my sleep data, suggests what I should eat based on my nutrition and activity, and gives me a clean, relevant overview of my day—appointments, important updates, maybe even things I would otherwise miss.
The strange part is: most of the data already exists. Apple Health, calendars, emails, feeds—it’s all there. But accessing it in a clean, unified way is still painful.
Everything is siloed, APIs are limited, and stitching it together in a meaningful, privacy-preserving way is harder than it should be.
On the model side, I feel like we’re getting close. With newer, faster approaches (including diffusion-style models), local inference is becoming realistic. Running something like this fully private—behind your own firewall—doesn’t sound crazy anymore.
You could even imagine it influencing things like DNS filtering or network rules dynamically. And honestly, even today, most of the pieces are already there. My Hermes agent setup with free models via OpenRouter can already orchestrate parts of this. The missing piece isn’t intelligence—it’s integration, memory, and reliable data access.
So maybe the real problem isn’t building a smarter assistant. Maybe it’s building a system that actually understands your life as a stream of events, not just prompts.
I’m curious how others are approaching this. Are you building assistants—or something that actually feels like a personal system?