Emma

Emma is a privacy-first, open-source AI assistant that runs entirely on your Mac. No cloud, no data leaks: everything stays local.

Main features
• Fully local: models run on-device using Apple Silicon (Metal) so chats, files and voice never leave your machine.
• Private by design: no sign-up, no telemetry, no remote logging; GitHub-verifiable code.
• Universal file brain: drop PDFs, DOCX, TXT, Markdown, images or code; Emma builds an instant, searchable knowledge base and cites exact pages.
• Context-aware chat: ask across all stored documents in one thread, follow up with natural language, receive concise answers plus source previews.
• Native Mac speed: 7-40 B parameter models quantized to 4-8 bit, leveraging Neural Engine & GPU for 2-3× faster inference than cloud equivalents.
• One-click model hub: browse, download and switch between Llama-3, Phi-3, Mistral, Gemma and more; automatic updates keep you current.
• Voice input & dictation: system-wide shortcut wakes Emma for speech-to-text queries without sending audio anywhere.
• Offline OCR & vision: extract text from scanned PDFs or images and query charts, screenshots or photos on the spot.
• Encryption & export: chat history is SQLite-encrypted; export threads as Markdown or JSON anytime.
• Free community edition: core app MIT-licensed; optional one-time Pro unlock adds larger model presets, priority support and dark-menu mode.

In short, Emma turns your Mac into a self-contained AI workspace where convenience, speed and complete privacy coexist.