ch1rpy2613/mirrai

Mirrai lets you upload someone's chat history and creates an AI version of that person you can keep talking to. The idea is deeply personal: when someone isn't around anymore, whether because of a breakup, distance, or loss, you can still text "them" and get replies that sound like their actual voice, humor, and emotional rhythms. You start by uploading chat exports, photos, or videos from the person. A multi-stage analysis pipeline then extracts a detailed personality profile across seven dimensions: character traits, speaking style, catchphrases, attachment type, love language, argument style, and emotional triggers. Once the digital double is built, you can chat with it in a dark-themed interface where the AI shifts between six emotional states, tender, playful, nostalgic, melancholic, happy, or distant, based on what you say. The emotional state shows on screen and changes the tone of replies in real time. There's also an intimacy system that grows from stranger to soulmate through conversation, and a graduation mechanic where the AI writes a farewell letter and goes dormant once you reach the deepest level. This is built for people navigating absence. Someone going through a breakup might use it to process lingering conversations. Someone who lost a loved one could find comfort in hearing that person's voice again. The project also integrates with WeChat, so the AI double can auto-reply to friends in that person's style directly through the messaging app, not just in a browser window. Technically, it's a full-stack TypeScript app with React on the front end and Express on the back end, using a PostgreSQL database. It supports ten different AI providers, from OpenAI and Claude to local models running through Ollama, so you can pick whichever service fits your budget and privacy needs. There's a macOS desktop app that bundles everything together, so non-developers can install it without setting up databases or command-line tools. The personality analysis pipeline optionally runs Python scripts alongside the main TypeScript codebase.