Descriptions:
Alpha Stack walks through building ‘Claudina,’ a 3D animated desktop companion that integrates with Claude Code’s MCP server to give an AI avatar real computer control — browsing the web, managing files, and executing tasks through natural voice conversation. The project is a hands-on demonstration of chaining multiple AI components into a single interactive system.
The tech stack combines Kokoro (local text-to-speech), OpenAI’s Whisper (speech-to-text), Three.js with the VRM SDK for rendering an animated 3D avatar, and Electron for the floating desktop window. A Python server handles all audio processing and model inference. The glue between the avatar and Claude Code is a mailbox-style message queue: voice input is transcribed, passed as a message, and Claude Code polls the queue and acts on it. The entire project is scaffolded from a 25,000+ token cheat sheet that Claude Code (running in Opus) uses to one-shot the full build with pre-approved permissions for Bash, Python, and Electron.
For users with CUDA-capable GPUs, the setup optionally supports Quen 3 TTS for fully customizable voice output, though the presenter warns it runs too slowly on laptop hardware. The video is an unpolished but honest builder’s walkthrough — including a live moment where Claudina refuses to open a browser despite being theoretically wired to do so — and is aimed at developers who want to extend or remix the concept for their own desktop AI integrations.
📺 Source: Alpha Stack · Published February 27, 2026
🏷️ Format: Hands On Build
