aq CLI

aq gpus is the one aq command that works before you've done anything else: no aq login, no account, no API key. Install the CLI and run it bare, and it prints a market-summary table (one row per GPU model, cheapest per-GPU rate first, with how many providers compete for it), the same feed behind the GPU price index.

Add --gpu (a substring match, so --gpu H100 works), --provider, --region, or --max-price, and it drops into a per-offer table listing every individual offer that matches; add --json to get those raw offers in a script-friendly form. That table prints two price columns: $/GPU-HR and $/HR TOTAL, because the underlying feed reports a whole-offer total for most providers and a flat per-GPU rate for one (Akash); aq gpus normalizes both, so $/GPU-HR is always the column to compare providers on.

The installer is a single script that resolves the latest release, verifies its SHA-256 checksum, and drops the binary in /usr/local/bin or ~/.local/bin. Nothing else to configure.

aq login pairs the CLI to your account with a device code: it prints a short code, you approve it once in the console, and the CLI holds a scoped credential from then on. No copying a token into a dotfile.

aq up --jupyter or aq up --comfyui finds the cheapest GPU matching your constraints, rents it, brings the app up, and prints the URL and credential. Narrow it with --gpu (a substring, so --gpu "RTX 4090" works), --max-price to cap the hourly rate, or --provider to pin one vendor.

If you have no SSH key registered yet, aq up registers a local one for you rather than failing with a lecture.

aq ssh manages a keypair and writes a real ssh_config host block, then hands your terminal to the actual ssh binary. Because the alias aq-<name> is a genuine Host entry, everything that speaks ssh_config just works: scp, rsync, and VSCode Remote-SSH included.

It also takes -- <cmd> to run something and exit, and -L to forward a port, the usual ssh ergonomics, because it is ssh.

aq deploy --snapshot <id> rents a new GPU and restores an existing workspace snapshot onto it in one command, optionally relaunching Jupyter or ComfyUI on the restored data, or --no-app to restore only.

The same --gpu, --max-price and --provider filters apply, which is the whole point: moving your environment to a cheaper GPU somewhere else is one command from a terminal.

aq import runs on a box you already rent somewhere else (RunPod, Vast, your own hardware) and captures it into a portable Aquanode setup. It's survey-first: before anything uploads, it prints exactly what it detected, what it's skipping (with real sizes, so a large dataset you forgot about doesn't just vanish), and what it couldn't even read: a third state, distinct from "skipped on purpose." --include and --exclude adjust the set before you commit.

It captures directories, not the machine: no apt state, no systemd units, no driver install get replayed. Once it's in, --launch rents a GPU and restores onto it in the same command, defaulting to the GPU model it observed on the source box.

aq status prints a deployment's state, its URL and its credentials (hidden unless you pass --show-secrets). aq down tears the box down and stops the meter. aq whoami and aq logout cover the credential itself.