Connecting via SSH

Generate a key, add it, connect, troubleshoot.

1. Generate a key (if you don’t have one)

macOS / Linux

ssh-keygen -t ed25519 -C "[e-mail]"

Windows (PowerShell)

ssh-keygen -t ed25519 -C "[e-mail]"

Accept the default location. Either skip the passphrase (faster) or set one (more secure — recommended for shared machines).

The command creates two files:

• ~/.ssh/id_ed25519 — your private key. Never share. Never paste anywhere.
• ~/.ssh/id_ed25519.pub — your public key. Safe to share.

2. Add the public key

macOS

cat ~/.ssh/id_ed25519.pub | pbcopy

Linux (with xclip)

cat ~/.ssh/id_ed25519.pub | xclip -selection clipboard

Windows PowerShell

Get-Content ~/.ssh/id_ed25519.pub | Set-Clipboard

Now in the dashboard: Settings → SSH keys → + Add key. Paste, give it a name (macbook-2026), save.

3. Connect to an instance

After launching a pod, the modal shows something like:

ssh root@123.45.67.89 -p 22001

Copy that, paste into your terminal:

ssh root@123.45.67.89 -p 22001
# The authenticity of host '[123.45.67.89]:22001' can't be established.
# ED25519 key fingerprint is SHA256:abc…
# Are you sure you want to continue connecting (yes/no/[fingerprint])? yes
# Welcome to Ubuntu 24.04 …
root@gpu-pod-01:~#

Confirm yes at the fingerprint prompt the first time.

4. Save the connection in your SSH config

If you’ll reconnect often:

~/.ssh/config

Host gpu-rtx4090
    HostName 123.45.67.89
    Port 22001
    User root
    IdentityFile ~/.ssh/id_ed25519
    StrictHostKeyChecking no
    UserKnownHostsFile=/dev/null

Then just ssh gpu-rtx4090. StrictHostKeyChecking no + the /dev/null known_hosts trick is reasonable here because every rental gets a fresh IP and you’d otherwise build up a giant known_hosts that you’d have to clear.

Troubleshooting

Permission denied (publickey)

Your local key isn’t matching any of the keys you’ve added. Verify:

ssh-add -L | grep ed25519

Should match the key you pasted in the dashboard. If it doesn’t, you may have multiple keys — pass -i explicitly:

ssh -i ~/.ssh/id_ed25519 root@host -p 22001

Connection timed out

Three causes:

1. Pod still provisioning — wait until status says running in the dashboard
2. Wrong port — every pod gets a random high port. Check the modal.
3. Your firewall — corporate VPNs sometimes block outbound on non-standard ports. Test from a different network.

Connection refused

Pod’s SSH daemon isn’t up yet — happens during the last 5 seconds of provisioning. Wait, retry.

My key worked yesterday, doesn't today

You probably deleted the key from the dashboard and launched a new pod. New pods only see the keys present at the moment of launch. Re-add the key, launch a fresh pod.

Multiple keys, want to use a specific one

Either use -i ~/.ssh/specific_key per-command or pin it in ~/.ssh/config per-host (see above).

Copying files

Local → pod (entire folder)

rsync -avz --progress ./project root@host:/root/ -e "ssh -p 22001"

Pod → local (single file)

scp -P 22001 root@host:/root/output.bin ./

Pod → S3 (from inside the pod)

aws s3 cp /root/model.safetensors s3://your-bucket/

Port forwarding (Jupyter, vLLM web UI, etc.)

If you’re running Jupyter on port 8888 inside the pod and want it on localhost:8888 locally:

ssh -L 8888:localhost:8888 root@host -p 22001

Then open http://localhost:8888 in your browser.