Can Claude start and babysit a training run on a remote machine?

Yes. Claude can launch the training command, poll progress and GPU utilization, read the logs for errors, and report back — all through the remote_exec tool against your GPU box's session code.

Developers & cloud

Drive a remote GPU & ML training box from Claude

Q: How do I monitor a remote GPU box without a screen?

Install the AI Commander agent on the GPU machine and ask your AI client to run nvidia-smi or read training logs by quoting the session code. The text output streams back, so a display is never needed — ideal for headless rigs and rented GPU instances.

Q: Does it work with rented GPUs and cloud instances?

Yes. Because the agent only needs outbound HTTPS, it runs on rented GPU instances, cloud VMs, and your own rig at home or in the office, even behind NAT.

Your GPU box lives in a closet, a colo, or a rented instance — headless, no monitor, often behind NAT. AI Commander lets Claude operate it over a real shell: start a training run, watch nvidia-smi, diagnose a CUDA mismatch, and pull the latest metrics, all from your laptop without exposing SSH.

The job

ML rigs are the definition of headless: there's nothing to screen-share, and everything happens at the command line. The friction is reaching them — port forwarding, jump hosts, or fiddly tunnels. AI Commander removes that and adds an agent that can reason about the output:

Launch a run and keep an eye on it: python train.py + periodic nvidia-smi.
Diagnose the classic "CUDA error: no kernel image" by checking driver/toolkit versions.
Free a stuck GPU: find the orphaned process and kill it.
Check disk before a checkpoint write, then tail the loss to see if it's converging.

What it looks like

Claude — MCP

You: on gpu-rig, is the training still running and how's GPU memory? Claude: > nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total --format=csv 96 %, 22360 MiB, 24564 MiB > tail -n 3 runs/exp42/train.log epoch 7/20 loss 0.214 val 0.231 Claude: Still training — epoch 7/20, GPU at 96% and 22.4/24.6 GB used. Loss is trending down.

Why AI Commander for GPU work

	AI Commander	SSH tunnel	Jupyter exposed	Cloud sandbox (E2B)
Your actual rig	✓	✓	✓	ephemeral
No inbound port	✓	✗	✗	✓
Headless / no display	✓	✓	browser	✓
AI client drives it	✓ MCP	✗	manual	SDK

Set it up

On the GPU machine (Linux), install the agent:

curl -fsSL https://aicommander.dev/install | sudo bash

Connect your AI client and quote the session code (or alias it gpu-rig). Then describe what you want done.

FAQ

How do I monitor a remote GPU box without a screen?

Install the agent on the GPU machine and ask your AI client to run nvidia-smi or read training logs by quoting the session code. The text output streams back, so a display is never needed — ideal for headless rigs and rented GPU instances.

Can Claude start and babysit a training run remotely?

Yes. Claude can launch the training command, poll progress and GPU utilization, read logs for errors, and report back — all through remote_exec against your GPU box's session code.

Does it work with rented GPUs and cloud instances?

Yes. The agent only needs outbound HTTPS, so it runs on rented GPU instances, cloud VMs, and your own rig — even behind NAT.

Talk to your GPU box

Install the agent and let Claude start, watch, and debug your training runs.

Install the agent Connect your AI client