Nyxe Flow — Mzed Studio case study

Native Mac app that transcribes meetings and dictation on-device, then drafts summaries and action items.

Overview

A native Mac app that captures meetings and dictation, transcribes and labels speakers on the device, and turns the result into summaries and follow-ups.

The problem

Meeting notes usually mean a bot in the call or audio uploaded to the cloud, then follow-ups written by hand. Flow keeps the audio on the Mac and does the follow-up work for you.

What we built

• On-device transcription on the Mac's Neural Engine, ready from first launch with no model downloads
• Captures both microphone and system audio, with speaker labels worked out on the device
• Summaries and action items on Apple's on-device models, with a switch that blocks all cloud processing
• Dictation into any text field from a global hotkey, with voice edit commands and a custom vocabulary
• Automatic meeting detection for native apps and browser-based calls
• Action items to Reminders or Calendar, follow-ups as Mail drafts, and agent access through a CLI and MCP with consent and an audit log

Stack

Swift 6, SwiftUI, CoreML, Apple Foundation Models, SQLite, MCP

Outcome

Live at nyxe.app/flow for Apple silicon Macs, included with Nyxe Pro.

Industry: macOS, On-device AI

Designed, built, and shipped by Mzed Studio.