Marven
DemoMarven is a macOS AI desktop assistant that blends conversational AI with real system actions. Built with Next.js, React, TypeScript, and Electron, it supports both cloud and local models via Groq and Ollama, voice interaction with wake-word detection, text-to-speech, screen understanding, and desktop automation including opening apps, managing media, screenshots, timers, and quick briefings.
Next.jsReactTypeScriptElectronGroqOllamaTailwind CSSSwift
- Problem
- Most AI assistants are browser-based chat tools with no real connection to your machine. There was a gap for a desktop-native assistant that could both converse intelligently and actually control your system through natural language.
- Solution
- Built a Next.js app embedded inside Electron with a polished multi-conversation chat UI, then layered in Groq for cloud inference and Ollama for fully local LLM support. Added a voice stack (microphone capture, Groq Whisper transcription, and macOS TTS) alongside natural-language desktop commands powered by child_process, osascript, and open. Extended with a native Swift menu-bar helper for global hotkey support.
- Challenges
- Bridging the browser-based Next.js frontend and Electron's main process for system-level commands required careful IPC design. Keeping voice capture responsive, Whisper transcription fast, and TTS playback smooth without blocking the UI was the trickiest part of the real-time pipeline.
- Outcome
- A fully working macOS AI assistant that handles streamed AI chat, voice I/O, and a wide range of desktop automations from a single interface. Demonstrates full-stack desktop development across React, TypeScript, Electron, native Swift, and multi-provider AI integration.