I've spent years being the person in the meeting who takes the notes. Not because I'm organized — because nobody else will, and if nobody writes anything down, the meeting might as well have never happened. The problem was never taking notes. It was that the notes I produced were a wall of text nobody actually read the next day.
Here's what a note from my old system looked like: one big jumbled paragraph with everyone's name, a couple of tangents, and maybe one real decision buried somewhere in the middle. Re-reading it meant re-interpreting it. Pulling the action items out meant scanning line by line. It was busywork, and the whole team treated the notes like junk mail.
When I finally got tired of it, I didn't sign up for another cloud meeting assistant. I already had the pieces sitting on my own machine: Whisper for transcription and LM Studio running a small local model. I just needed something to glue them together. The result is a Windows app that records the meeting, transcribes it locally, and hands the text to a local LLM that hands back actual structure: a Summary, Key Decisions, Action Items (with an owner and date where someone actually stated them), and Open Questions. It exports to Markdown, plain text, or PDF.
The weirdly nice part is the Action Items. Instead of "we should follow up with the vendor" buried two paragraphs deep, you get a clean line that says who owns it and roughly when. When someone in my team asks what they agreed to on Thursday, it's right there. The meeting ends and the minutes are already done — I'm not scrubbing through a recording at 5pm trying to remember who said what.
It's all local. No cloud, no account, no subscription, nothing uploaded. That matters to me because roughly half the meetings I sit in touch on real infrastructure — vendors, access, security decisions. I am not comfortable shipping that transcript to some vendor's storage so a stranger can train on it. The app needs Windows 10/11, Python 3.10+, and LM Studio with any small model loaded. It runs fine on CPU — a GPU just makes transcription faster, it's not required. Under the hood it's faster-whisper for the audio, and it grabs system audio through WASAPI loopback, which means it catches what you're hearing from Teams or Meet even when you're not the one talking — not just your own mic.
One honest caveat: recording meetings can be regulated where you work, and in a lot of places it requires consent from everyone in the room, especially when you're capturing other people's audio. Check your policies and tell people. This app doesn't dodge that for you.
If your notes look like mine used to and you'd rather keep them off someone else's servers, it's a one-time $9 at https://symshah.gumroad.com/l/xgoypx. No subscription — the subscription model is literally why I built my own thing in the first place.
Top comments (0)