Mirrai: AI Digital Persona Generation

Mirrai creates AI digital personas from user data, primarily chat logs, to address emotional distance, relationship anxiety, and processing loss. It generates a sophisticated replica of an individual, offering a persistent digital presence for connection, emotional processing, or facilitating closure.

How It Works

A multi-stage personality distillation engine deeply analyzes uploaded data to construct personas across seven structural layers (traits, speech, triggers, etc.), aiming for authentic replication. A dynamic emotional state machine (six states) adapts responses in real-time, while a unified LLM abstraction layer supports over ten AI providers for flexible and optimal performance.

Quick Start & Requirements

Setup involves cloning the repo, installing Node.js dependencies (pnpm install), and configuring PostgreSQL 14+. Prerequisites include Node.js 20+, pnpm/npm, and Python 3.9+ (for the skill engine). Environment variables for database and LLM API keys are essential. A macOS one-click installer is also provided.

Highlighted Details

• Multi-stage Personality Distillation: Generates personas with 7 structural layers for deep replication.
• Dynamic Emotional State Machine: Simulates 6 adaptive emotional states (Gentle, Playful, Nostalgic, Melancholy, Happy, Aloof).
• Unified LLM Abstraction: Integrates 10+ AI providers (OpenAI, Claude, etc.) via a single interface.
• WeChat Bot Integration: Uses Wechaty for personas to auto-reply to WeChat messages.
• Full-Stack Type Safety: Employs TypeScript and tRPC for end-to-end type checking.

Maintenance & Community

The project is primarily maintained by the author "OpenDemon." Specific community channels or a public roadmap are not detailed in the provided README.

Licensing & Compatibility

Released under the MIT License, permitting broad use, modification, and distribution, including for commercial purposes.

Limitations & Caveats

Persona quality hinges on input data quantity and quality. Optimal AI responses require powerful LLMs (GPT-4o, Claude recommended). The WeChat bot needs manual QR code scanning for authentication. The project features a "graduation mechanism," indicating an intended persona lifecycle.