read the last line of my paragarph - "If you ever had that moment of losing track of files or just wanted actual control over your laptop without some company in the cloud watching what you're doing... might be worth checking out."
39 karma · joined February 5, 2026
It runs via Ollama on the same machine as the assistant. There’s no cloud inference, no API calls to OpenAI/Anthropic/etc., and no data sent out for processing. The model is only used to parse intent and time expressions, not to execute actions or access credentials.
If the machine has no internet connection, the assistant still works locally, the only thing you lose is the ability to message it remotely via Telegram.
Happy to clarify further, but this is intentionally designed to be local-first end to end.
If people with macOS or Linux experience have thoughts on architecture or portability tradeoffs, I’m happy to discuss and compare approaches.
Forgot the name. Forgot where I saved it. Just remembered it was something I read for hours one evening.
That happens to everyone right?
So I thought - why can't I just tell my computer "send me that PDF I was reading 5 days ago at evening" and get it back in seconds?
That's when I started building ZYRON. I am not going to talk about the development & programming part, that's already in my Github.
Look, Microsoft has all these automation features. Google has them. Everyone has them. But here's the thing - your data goes to their servers. You're basically trading your privacy for convenience. Not for me.
I wanted something that stays on my laptop. Completely local. No cloud. No sending my file history to OpenAI or anyone else. Just me and my machine.
So I grabbed Ollama, installed the Qwen2.5-Coder 7B model in my laptop, connected it to my Telegram bot. Even runs smoothly on an 8GB RAM laptop - no need for some high-end LLMs. Basically, I'm just chatting with my laptop now from anywhere, anytime. Long as the laptop/desktop is on and connected to my home wifi , I can control it from outside. Text it from my phone "send me the file I was working on yesterday evening" and boom - there it is in seconds. No searching. No frustration.
Then I got thinking... why just files?
Added camera on/off control. Battery check. RAM, CPU, GPU status. Audio recording control. Screenshots. What apps are open right now. Then I did clipboard history sync - the thing Apple does between their devices but for Windows-to-Android. Copy something on my laptop, pull it up on my phone through the bot. Didn't see that anywhere else.
After that I think about browsers.
Built a Chromium extension. Works on Chrome, Brave, Edge, anything Chromium. Can see all my open tabs with links straight from my phone. Someone steals my laptop and clears the history? Doesn't matter. I still have it. Everything stays on my phone.
Is it finished? Nah. Still finding new stuff to throw in whenever I think of something useful.
But the whole point is - a personal AI that actually cares about your privacy because it never leaves your house.
It's open source. Check it out on GitHub if you want.
And before you ask - no, it's not some bloated desktop app sitting on your taskbar killing your battery. Runs completely in the background. Minimal energy. You won't even know it's there.
If you ever had that moment of losing track of files or just wanted actual control over your laptop without some company in the cloud watching what you're doing... might be worth checking out.
First: marginal inference cost vs total business profitability. It’s very plausible (and increasingly likely) that OpenAI/Anthropic are profitable on a per-token marginal basis, especially given how cheap equivalent open-weight inference has become. Third-party providers are effectively price-discovering the floor for inference.
Second: model lifecycle economics. Training costs are lumpy, front-loaded, and hard to amortize cleanly. Even if inference margins are positive today, the question is whether those margins are sufficient to pay off the training run before the model is obsoleted by the next release. That’s a very different problem than “are they losing money per request”.
Both sides here can be right at the same time: inference can be profitable, while the overall model program is still underwater. Benchmarks and pricing debates don’t really settle that, because they ignore cadence and depreciation.
IMO the interesting question isn’t “are they subsidizing inference?” but “how long does a frontier model need to stay competitive for the economics to close?”