MeetWise

TL;DR

• A desktop meeting copilot: a live transcript and the screen you’re sharing feed one short suggestion at a time into a frameless, always-on-top overlay.
• After the call it turns the transcript into a summary, decisions, action items and open questions, and drafts the follow-up email.
• It runs on any OpenAI-compatible endpoint (vision included) or fully locally through Ollama.

What it does

The live-session pipeline pairs a Web Speech API microphone transcript with screen sampling at one frame per second, keeping a snapshot every 8 seconds. When the transcript moves, it waits 1.4 seconds, then asks the model for one coaching line: what to say next, or a question worth asking.

On the Electron side: a tray menu, IPC between the main window and the overlay, and a privileged app:// scheme that gives speech recognition the secure context it needs.

A mock fallback when no model is configured → the UI always runs, so you can build and demo it offline → you have to watch for mock output being mistaken for real output.

Status

Built: the renderer compiles to a production bundle. It isn’t packaged as an installer, and live speech-to-text inside Electron is still unproven. Before any demo I’d add a screen picker; today it grants the first screen without asking.