OpenMeet - The meeting copilot that runs on your Mac
by•
Your calls stay on your Mac.
Most meeting tools work by uploading your conversation to a server. If you talk to clients under an NDA, handle anything regulated, or would simply rather not post your sales calls to a third party, that is a problem no feature list fixes.
OpenMeet transcribes on device. Speech recognition, speaker separation, and answer drafting all run locally. Nothing about your call is uploaded, because there is nowhere for it to go
Replies
Hey Hunters 👋
I'm Selçuk. OpenMeet is a menu-bar app for macOS that listens to both sides of a call, notices when you've been asked something, and drafts an answer while the other person is still talking.
It started as a folder of Python scripts I wrote to get through my own technical interviews. I interview in English, which is not my first language, and the gap that kept costing me wasn't knowledge — it was the four or five seconds between hearing a question and having the English for it. So I wired Deepgram to a local model and read the drafts off a second window.
Most of what I learned in that folder was unglamorous. I keep a list of terms the transcriber has to be told about, because "pod" comes back as "port" and "node" comes back as "not", and an answer built on a misheard noun is worse than no answer. I had to forbid the model from writing YAML or CLI commands, because I was reading the output out loud and you cannot say a Helm chart. I had to forbid it from saying "you'd check the kubelet" — the interviewer heard me instruct them instead of answering. And for a while the prompt hard-coded "senior DevOps interview", so "tell me about a time you disagreed with your manager" came back as a paragraph about pod counts. The field now comes from the role and your own resume, not from me.
OpenMeet is that folder, turned into something other people can install.
What it does during a call
Hears both sides separately. Your microphone and the audio the other side arrives on are captured on two channels and transcribed apart, so a panel of two reads as THEM 1 and THEM 2 instead of collapsing into one voice.
Spots the question. It watches the far side for questions — English and Turkish — and starts drafting without you pressing anything.
Drafts from your own notes. Point it at a folder and it ranks across it for what's relevant, then names the files it used at the bottom of the draft. If nothing was relevant it says no notes used rather than implying a source.
Writes answers you can actually say. First person, no code blocks, no pasted commands unless you're asked for one. That constraint is the whole difference between a prompter and a wiki.
Rewrites on one key. Shorter, give an example, name the trade-off, or tell me what to ask back.
Stays out of the screen share. Its windows are excluded from capture at the window-server level, so it isn't in the frames macOS hands to Zoom, Meet, Teams, QuickTime or OBS