Put your machines to work
Ask your AI client to manage a home server, run a training job or investigate a production issue. Explore examples for the hardware you already own.
nvidia-smi, fix a CUDA error, free a stuck card, or grab the loss curve.Claude Code · Codex · ChatGPT harness
Not a replacement for those tools — the remote-machine layer they plug into over MCP.
Grok CLI / grok build harness
Keep Grok. Give it your machines over MCP — remote shell, jobs, and file transfer. Not a replacement.
Cursor MCP harness
Keep Cursor. Give it your machines over MCP — remote shell, jobs, and file transfer. Not a replacement.
Codex CLI harness
Keep Codex. Give it your machines over MCP — remote shell, jobs, and file transfer. Not a replacement.
ChatGPT / OpenAI harness
Keep ChatGPT, GPT, and submodels like Astra (gpt-6-astra). Give them your machines over MCP. Not a replacement.
claude -p run on a box with no operator present — a non-root, allowlisted service token instead of an interactive code.One setup for every example
Install the agent and connect your AI client. The machine connects outbound, so you do not need inbound ports or a VPN. Follow the setup guide.
To collect a report, checkpoint or rendered image, or send a config file, add file transfer with Pro.
Pick a machine and try it
Install the agent on one Linux box, or run the desktop app on a Mac or PC, and ask your AI client to do something on it. Fresh codes work for the first hour without an account; sign in for ongoing access.