Built an offline AI meeting-notes app that runs on CPU, no GPU. Then found out how it stacks up against free alternatives.
A client of mine once asked me to just email over meeting notes afterward. I said sure, then realized the transcript was sitting in some SaaS tool's cloud dashboard, and I had no idea where it actually lived. That's the moment I decided to build my own instead of trusting whatever was cheapest.
What it actually does: paste in a transcript you already have (from Zoom, Teams, Otter, wherever), pick an output format, get notes back in seconds, entirely on your machine:
Full notes : complete structured breakdown
Quick summary : 5 bullet points, key takeaways only
Action items only : who owns what, by when
Email-ready : formatted to paste straight into an email to your team
Runs on a small local model ( under 1GB) , 6GB RAM minimum, no GPU, ~2GB disk for one-time setup. Free tier: 3 transcripts/day, no account needed. Pro: $10/mo, Team: $19/mo for 5 seats payment processing is stuck in approval limbo right now, so it's a manual waitlist-and-follow-up in the meantime.
[60-second demo here, if you want to see it working before reading further: link]
Posted about this on Reddit and got pointed to Meetily and OpenWhispr , both solid, more mature, open-source, and further along in one real way: they record meetings live, mine only works from an existing transcript. The one place I've found a genuine, checkable difference: they lean on Ollama with 7B+ models for good output, which realistically wants ~16GB RAM and a GPU. Mine's built to run on a normal 5-8 year old laptop, not a machine spec'd for AI work.
Solo dev, India-based, still very early, zero real paying customers. One person tried the free tier off Reddit and came back saying it looked solid, which is the first real signal so far. Curious what this community thinks: is "runs on modest hardware" enough of a wedge against a free, more mature open-source competitor, or does a bootstrapped solo dev need a sharper angle entirely?
