Yoji is a privacy-first AI desktop companion designed to offer advanced OS-level control and a dynamic, emotional personality. It targets power users and developers seeking a highly customizable, locally-processed assistant that can manage files, run scripts, interact with web services via an open protocol (MCP), and evolve its behavior through user interaction and custom sub-agents. The primary benefit is a powerful, extensible AI that respects user privacy by keeping all data and processing on the local machine.

How It Works

Yoji operates on a local-first, privacy-centric architecture, eschewing cloud dependencies. Its core functionality revolves around direct OS interaction, enabling sophisticated file management, script execution, and web searches. Extensibility is achieved through the Model Context Protocol (MCP), an open standard allowing seamless integration of third-party services and tools, even enabling the AI to discover and install them. A key differentiator is its simulated "emotion engine," which uses neurotransmitter models to influence response style and UI, creating a more human-like interaction. A multi-layered memory system ensures persistent learning and personalization, all stored locally.

Quick Start & Requirements

• Primary Install/Run: Clone the repository, initialize submodules (git submodule update --init), install dependencies with pnpm install, and run with pnpm dev electron.
• Prerequisites: pnpm (v11 recommended), git. API keys for models like DeepSeek or Qwen are required post-setup.
• Dependencies: macOS native speech recognition is required for voice wake-up and TTS features.
• Links: GitHub Repository

Highlighted Details

• Local-First & Privacy: All data, including memory, configuration, and processing, remains on the user's machine.
• MCP Extensibility: An open protocol allows for theoretically unlimited integration of external services and tools, managed dynamically.
• Advanced Sub-Agent System: A local factory and scheduling system supports both synchronous (immediate) and asynchronous (background) sub-agents for complex task orchestration.
• Simulated Emotion Engine: A core component driving nuanced, human-like responses and UI changes based on simulated neurotransmitters.
• Intelligent Document Handling: Direct parsing and text extraction from PDF and DOCX files.
• Voice Wake-up & TTS: Full offline voice interaction capabilities are available on macOS.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels like Discord or Slack. The focus is on the technical implementation and features.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use. The core application supports macOS, Windows, and Linux, but advanced features like voice wake-up and TTS are currently macOS-exclusive.

Limitations & Caveats

Certain advanced features, specifically voice wake-up and text-to-speech, are currently limited to macOS. The fully autonomous mode for proactive AI chatting is not exposed externally due to operational cost considerations. Functionality relies on user-provided API keys for external LLM models, implying potential third-party service costs.