ghdtjrtka/ghost-vessel

Ghost Vessel is a desktop app that gives your local AI agent a face and a body to live in. Instead of talking to your agent through a plain text chat window, you get a small, always present avatar on your screen that looks like a video call. The avatar reacts to what the agent says, showing facial expressions, blinking, and breathing during idle moments. The project calls the agent the "ghost" and the avatar the "vessel" it gets bound into. It works with agents like Hermes and OpenClaw, or anything else that can connect. The app splits the agent's replies into three parts. One part handles the visual performance, the emotion the avatar shows on its face. Another part is the spoken dialogue, played through a local text to speech voice. The third part is data like code and files, which show up as cards in a chat panel and never get read aloud. The avatar also has a mood system. If you scold it, its mood drops and it rests in a subdued state. If you praise it, it brightens. That mood state persists across restarts. A notable feature is that the avatar window serves your agent's actual slash command menu. If your agent has commands like /model or /status in a messenger app, the same list appears here. You type /, the menu appears, and the commands pass straight through to the agent. The system is agent agnostic, with adapters for Hermes and OpenClaw plus a demo responder that needs no setup. Voices are swappable, with local and cloud options, and you can pick them in settings. Avatars are pure data bundles. You get a free 2D starter avatar called Haneul when you clone the repo, so it runs immediately. A photoreal avatar named Yeoreum is sold separately. You can build your own avatar by creating one front facing image, animating it into short video clips with an image to video model, and dropping the clips into a folder. The engine plays the clips and does not care what art style produced them. The full README is longer than what was shown.