vmux in action.
Install the vmux skill for Claude Code or Codex. Then just ask: "train nanogpt on an H100" — the agent spins up GPU, clones repo, trains.
Full-stack LLM chat with vLLM backend and Next.js frontend. Model downloads, loads, and serves — all in one command.
JupyterLab in the cloud with GPU access. Torch, JAX, matplotlib pre-installed. No password, accessible via preview URL.
Deploy FastAPI, Gradio, Streamlit instantly. Get a public URL. WebSockets work. Share with anyone.