SkyPilot.
Drop cloud: sixtytwo into your resources block and your task runs on our GPUs at the prices you see in the catalog, billed per minute.
Step 1
Install the adapter
Linux or macOS. SkyPilot itself doesn't support Windows clients.
pip install "sixtytwo-cli[skypilot]"
Step 2
Create an API token
On your account
page, generate a sixtytwo API token and save it to
~/.sixtytwo/credentials:
mkdir -p ~/.sixtytwo echo "token=sxty_..." > ~/.sixtytwo/credentials chmod 600 ~/.sixtytwo/credentialsOr export
SIXTYTWO_TOKEN=sxty_... in your shell.
Step 3
Verify with
sky check
sky check # Sixtytwo: enabled [compute] sky show-gpus --infra sixtytwo # Lists the same SKUs you see on sixtytwo.ai/, at the same prices.
Step 4
Point a task at sixtytwo
Standard SkyPilot YAML. Change one line.
resources: cloud: sixtytwo accelerators: A100-80GB:3 use_spot: false # spot tier costs roughly half run: | python train.py
sky launch train.yaml
sky logs, sky status, sky down all work.
Step 5 (optional)
Join your tailnet at boot
Need pods to reach internal services (MLFlow, feature stores,
GCS proxies) inside your VPC without a per-cloud VPN?
Save a Tailscale
auth key on your account, then add a setup block to your YAML:
resources: cloud: sixtytwo accelerators: A100-80GB:3 setup: | # Pull TS_AUTHKEY + reservation id from PID 1's env (providers # expose spec.env there, not in /etc/environment). export TS_AUTHKEY=$(tr '\0' '\n' < /proc/1/environ | grep '^TS_AUTHKEY=' | cut -d= -f2-) export SIXTYTWO_RESERVATION_ID=$(tr '\0' '\n' < /proc/1/environ | grep '^SIXTYTWO_RESERVATION_ID=' | cut -d= -f2-) # Tailscale hostnames must be DNS labels; rsv_xxx has underscores. HOST=sixtytwo-$(echo "$SIXTYTWO_RESERVATION_ID" | tr "_" "-") # Install Tailscale. curl -fsSL https://tailscale.com/install.sh | sh # Pod images don't ship systemd; start tailscaled in userspace-networking mode. [ "$(id -u)" = "0" ] && SUDO="" || SUDO="sudo -E" $SUDO tailscaled --tun=userspace-networking --state=/tmp/tailscaled.state >/tmp/tailscaled.log 2>&1 & sleep 2 $SUDO tailscale up --authkey=$TS_AUTHKEY --hostname=$HOST --accept-routes run: | python train.py # mlflow.example.tailnet.ts.net resolvesEach pod gets a tailnet hostname keyed to the reservation ID.
Spot, multi-node, and managed jobs.
Single-node spot works today, and preemption refunds the unused
tail. Multi-node and
sky jobs launch with managed
recovery aren't wired up yet. Email
founders@sixtytwo.ai if
you need them now.