Requires Python 3.10+
bash$ uv tool install jarvislabsCreate, pause, resume, and destroy GPU instances.
bash$ jl create --gpu A100Connect via SSH or copy files to and from your instance.
bash$ jl ssh <id>
jl upload <id> ./data /home/dataExpose ports for Gradio, FastAPI, or any web service.
bash$ jl create --gpu A100 --http-ports 8080,3000Your data survives pause/resume cycles automatically.
bash$ jl create --gpu A100 --storage 100Run scripts automatically when your instance starts.
bash$ jl create --gpu A100 --script-id <id>Deploy in India or Europe. More regions coming.
bash$ jl create --gpu H100 --region EU1Upload code, install deps, run your script, and stream logs — all in a single command. The instance auto-pauses when done.
Your local directory is synced to the instance.
Isolated Python environment is set up.
requirements.txt installed automatically.
Your training script executes on the GPU.
Live stdout/stderr streamed to your terminal.
Instance pauses when the script finishes.
bash$ jl run . --gpu A100 --script train.pyLet Claude Code, Cursor, or Codex drive your GPU experiments. Install agent skills with a single command.
$ jl setup ✓ API key configured ✓ Claude Code skills installed ✓ Default GPU: A100 $ # Now Claude Code can: $ jl create --gpu A100 ⠋ Provisioning instance... ✓ Instance ready in 38s $ jl run train.py --on <id> ✓ Training started — streaming logs... Epoch 1/10 loss=2.34 lr=5e-4 Epoch 2/10 loss=1.87 lr=5e-4 ...
One command sets up your API key, configures default GPU, and installs agent integration files.
bash$ jl setupInstall the CLI, grab an API key, and launch your first GPU instance in under two minutes.