# Marty

> Marty is an autonomous, Jarvis-class AI desktop companion designed natively for macOS (Apple Silicon M1-M4 and Intel). It floats seamlessly over your desktop, perceives what is on your screen in real time via Apple ScreenCaptureKit, converses with sub-100ms bidirectional voice via Google Gemini Live, and performs multi-step autonomous workflows across native Mac apps hands-free.

Marty operates on a 100% free Bring-Your-Own-Key (BYOK) model using Google AI Studio Gemini API keys (supporting Gemini 3.7 Flash, 3.5 Flash-Lite, and 3.1 Pro), providing zero subscription fees and complete local privacy.

## Quick Links

- [Official Website](https://www.martys.space/): Main website and download hub
- [Features Breakdown](https://www.martys.space/features): Complete breakdown of vision, voice, automation, and memory capabilities
- [How to Use Guide](https://www.martys.space/how-to-use): Step-by-step setup, permissions guide, shortcuts, and voice prompts
- [Frequently Asked Questions](https://www.martys.space/#faq): Common technical questions, pricing, and system requirements
- [Trust & Architecture](https://www.martys.space/trust): Security model, on-device encryption, and zero-training policy
- [Privacy Policy](https://www.martys.space/privacy): Privacy commitments and transparent data handling
- [Early Access Waitlist](https://www.martys.space/early-access): Weekly .dmg installer invite requests
- [Full Documentation](https://www.martys.space/llms-full.txt): Complete plain text technical knowledge base for LLM search engines

## Core Architecture & Capabilities

- **Continuous Bidirectional Voice**: Sub-100ms streaming latency powered by Google Gemini Live API. Uses Apple VoiceProcessingIO for hardware Acoustic Echo Cancellation (AEC), enabling natural interruption and barge-in. Offers 7 distinct voice personas (Capella, Puck, Charon, Kore, Fenrir, Aoede, Pegasus).
- **Live Screen Perception**: Continuous 1 FPS 1280x720 screen capture via Apple ScreenCaptureKit. Features native crosshair region snip tool and on-device Apple Vision OCR to analyze code, error dialogs, spreadsheets, and PDFs without manual screenshots.
- **Autonomous Multi-App Agentic Orchestration**: Chains sequential actions across Google Chrome, Apple Safari, Apple Mail, Apple Notes, Apple Pages, and system settings in a single instruction.
- **Semantic Browser Agents**: Parses live DOM in Chrome and Safari into structured outlines and interactive numbered badges [1], [2], [3] for precise clicking, typing, and form autofill.
- **Silent Background Research**: Queries DuckDuckGo and Wikipedia headlessly to fact-check and synthesize answers without opening disruptive browser windows.
- **Tri-Tier Local Memory Engine**: In-memory 6-turn conversational buffer, declarative key-value facts on disk, and long-term semantic vector memory using Apple NLEmbedding and Accelerate cosine similarity with a built-in Memory Inspector.
- **Privacy & Safety First**: 100% on-device wake word ("Hey Marty") via Speech.framework, automatic blacklisting of sensitive apps, confirmation gates before sending emails or destructive actions, and zero shell execution risks.

## System Requirements

- **Operating System**: macOS 14.0 Sonoma, macOS 15.0 Sequoia, or later
- **Hardware**: Apple Silicon (M1, M2, M3, M4, Pro, Max, Ultra) with Neural Engine optimization, or modern Intel Macs
- **API Key**: Free Google Gemini API key from Google AI Studio (aistudio.google.com)
- **Supported Models**: Gemini 3.7 Flash (Default), Gemini 3.5 Flash-Lite, Gemini 3.1 Pro
- **Cost**: 100% Free software with zero subscription fees (BYOK model)