Meeting Intelligence Copilot

01 The problem

A hard call moves fast. You are listening, thinking, and choosing your words at the same time, with no one feeding you the right line. You used to do all of that on your own. The right number is somewhere in a file you cannot open mid-sentence, and the perfect answer arrives ten minutes after you hang up. Notes written after the fact do not help in the moment. By then the call is over and the chance is gone.

02 The system

A desktop copilot that sits over the call. It hears your microphone and the other person’s audio at once and writes a live transcript as both of you talk. Press a hotkey and it reads the whole conversation, then streams two replies you can say out loud in seconds. Feed it a deck or your notes first and every line pulls from your real numbers, so you answer with the actual figure instead of a guess. When the call ends, one more hotkey writes a clean summary with action items and saves the full transcript to your notes. The work you used to do alone, and the follow-up you used to scribble afterward, both happen for you while the call is still live.

03 How it holds up

Raw speech runs through a local engine on the machine, so the app does not upload the meeting audio for transcription. The next privacy boundary is separate: when you request cloud suggestions or a summary, transcript text and selected deck or note context go to the configured provider. That distinction lets the operator decide whether the meeting is suitable for cloud processing and which provider account to use. The suggestion layer rejects clichés and repeats, and 117 tests cover the desktop workflow.

04 The result

A live call can carry a two-sided transcript, grounded reply suggestions, and a structured meeting note in one desktop workflow. Raw audio transcription stays local. Cloud suggestions and summaries use transcript text and selected context only when the operator invokes those features with a configured provider.

05 Frequently asked questions

What meeting data stays local and what goes to a cloud provider?

Raw meeting audio is transcribed locally on the laptop. When you request a cloud suggestion or summary, transcript text and selected deck or note context are sent to the configured provider. Sensitive calls should use only processing choices approved for that conversation.

How does the copilot hear both sides of a call?

It listens to your microphone and to the other person’s audio at the same time, and writes both into one live transcript as you talk. When you press the hotkey, it reads that whole two-sided conversation, so the line it hands you fits what was actually said.

How does the suggestion layer keep replies concise and specific?

The suggestion layer filters clichés and repeated lines before a reply reaches the screen. Each hotkey request sends the current transcript and selected context to the configured provider, then streams two grounded options for the user to assess.

What do you get when the meeting ends?

One more hotkey writes a clean summary with action items and saves it, along with the full transcript, to your notes. The follow-up you used to scribble after the call is written before you close the laptop, while the details are still fresh.