Command-line tool (cnx)

cnx launches, lists, connects to and stops your GPUs from a terminal, a script or CI. It comes with the Python SDK and uses the same API key.

Install and sign in

pip install computenexa          # or: pipx install computenexa
export CNX_API_KEY="cnxk_..."
cnx whoami

Create the key in the dashboard under API keys (see Use the API): "Look only" for listing, "Launch and stop" to launch.

A first session

cnx gpus --app datascience                       # pick a GPU
cnx launch rtx-4090 --app datascience --hours 2  # shows the cost, asks, then waits
cnx ssh cnx-7f3a9c                               # work on it
cnx stop cnx-7f3a9c                              # unused time refunded

Commands

Command	What it does	Example
`cnx whoami`	Who the key acts as, your team, its balance and GPU limit	`cnx whoami`
`cnx gpus`	GPUs for sale now, cheapest first (--app, --region us\|ca, --all)	`cnx gpus --app chat`
`cnx apps`	The apps a GPU can start with, and whether they take a model or need an SSH key	`cnx apps`
`cnx launch`	Launch a GPU; shows the cost and asks first (--yes to skip), waits until it's running	`cnx launch l40s --app datascience --hours 4`
`cnx ls`	Your running machines (--all includes ended ones; --everyone the whole team's)	`cnx ls`
`cnx show`	One machine: state, SSH command, API address, times and cost	`cnx show cnx-7f3a9c`
`cnx ssh`	Connect to a running machine; anything after the id is run on the machine	`cnx ssh cnx-7f3a9c nvidia-smi`
`cnx extend`	Add time at the same price	`cnx extend cnx-7f3a9c --hours 1`
`cnx stop`	Stop one or more machines; unused time is refunded	`cnx stop cnx-7f3a9c`
`cnx projects`	List projects; create or rm one	`cnx projects create titanic`
`cnx keys`	List SSH keys; add one (the public .pub file)	`cnx keys add "Work laptop" ~/.ssh/id_ed25519.pub`
`cnx usage`	GPU use and spending (--days)	`cnx usage --days 7`
`cnx receipts`	Finished sessions and their cost (--days)	`cnx receipts`
`cnx schedules`	List schedules; pause (--until), resume or rm one	`cnx schedules pause Weekdays --until 2026-12-01T08:00:00-08:00`

cnx --help and cnx <command> --help list every option.

cnx launch options

Option	Meaning
`gpu`	A GPU from cnx gpus, e.g. rtx-4090 (required)
`--app`	basic, datascience, chat, imagegen, modelapi or voice (required)
`--hours / --minutes`	How long to book (one of them, required); up to 7 days
`--project NAME`	Restore a project into ~/work and save it back; --create-project makes it if missing
`--label`	A name for the machine
`--model`	For chat, imagegen, modelapi and voice; default: the recommended model for the GPU
`--count`	GPUs in the machine (default 1)
`--region`	any (default), us or ca
`--no-wait / --timeout`	Return straight away, or how long to wait for it to be running (default 900 s)
`--yes`	Don't ask before spending (scripts and CI)

Scripts and CI

--yes: launches and stops don't ask. Without it, a script that isn't running in a terminal is refused rather than spending money silently.
--json on any command prints machine-readable output (the API's own fields).
Exit codes: 0 done, 1 refused (the reason and its request id on stderr), 2 wrong usage.
SSH from CI: add the CI's public SSH key once (cnx keys add), and accept a new machine's host key on first connect, as below.

# .github/workflows/train.yml (excerpt): train on a private GPU, then stop it
- run: pip install computenexa
- env:
    CNX_API_KEY: ${{ secrets.CNX_API_KEY }}      # a "launch and stop" key, stored as a secret
  run: |
    ID=$(cnx launch rtx-4090 --app datascience --hours 2 --project training --yes --json | jq -r .id)
    ssh -o StrictHostKeyChecking=accept-new "$ID@ssh.computenexa.com" "cd work && python train.py"
    cnx stop "$ID" --yes