Vultr + SSH setup
A Vultr instance gives you a GPU host you keep for the session. Studio can log in over SSH, install Docker and the NVIDIA container toolkit, pull the image and run the combined Studio + Compute container. No terminal required.
Sign-up link: it supports the project and does not change the price.
Before you start
- A Vultr account with billing enabled.
- SharpMind Studio running on this PC, with the server image pushed to Docker Hub (or use integral2u/sharpmind-server:gpu).
- A GPU plan and a region near you. The tested plan is vcg-a16-6c-60g-16vram.
- An SSH key. Studio can create one for you, and adding it to your Vultr account does not need an instance.
- For Deploy from the API, the key must already exist in your Vultr account; Studio never uploads a key.
Create the instance
- Deploy a Cloud GPU instance. Pick Ubuntu 24.04 and a GPU image that already includes the NVIDIA driver. A plain Ubuntu image works too, but you must install the driver and reboot before GPU jobs can use the card.
- Add the SSH key at creation. This is the easiest moment to attach it. If you do not have one yet, open Studio first and use Edit → Create key; the public key is shown there.
- Open the ports. Allow 22 for setup, 8080 for the Compute API and 5117 for the remote Studio UI. Restrict them to your IP where possible. A firewall that only allows 22 will make the host card time out even while the container is healthy.
- Keep the data folder. The image uses /data/SharpMind on the instance. If you create a separate block volume, mount it there; otherwise the instance disk is fine for a one-day run.
Add an SSH key
Vultr SSH keys belong to the account, not to an instance, so add one once before you deploy. Adding a key does not start or need an instance. Studio creates the private half on this PC and never uploads anything to your account.
-
Create or pick the key in Studio.
In Settings → Providers → Compute → Edit on the
Vultr host, the private key row offers:
- Create key generates ~/.ssh/sharpmind_ed25519 (ed25519, no passphrase) and shows the public key.
- Browse selects an existing OpenSSH/PEM private key instead.
- Copy public key puts the full line on the clipboard.
- Add the public key to Vultr. Open console.vultr.com/sshkeys, or in the console use the account menu → Account → SSH Keys (under Other) → Add SSH Key. Name it sharpmind-studio and paste the whole line, including the trailing comment: ssh-ed25519 AAAA… sharpmind-studio.
- Select it in Studio. Press Load plans on the Vultr panel and pick the key under Vultr SSH key. Deploy stays disabled until a key is selected and the matching private key is set on the host.
mkdir -p /root/.ssh
echo 'ssh-ed25519 AAAA...your-public-key... sharpmind-studio' \
>> /root/.ssh/authorized_keys
chmod 700 /root/.ssh
chmod 600 /root/.ssh/authorized_keys
The private key never leaves this PC. The host key is trusted on first use; this is a convenience for a fresh cloud box, not a substitute for pinning if you need it.
Add the host in Studio
- Add host → Vultr cloud server. Base URL http://<public-ip>:8080; Studio URL http://<public-ip>:5117.
- Token. Press Generate. The same value is written into the container as SHARPMIND_TOKEN, so Studio and the host match.
- Server image. Use the full registry name, for example integral2u/sharpmind-server:gpu or your-name/sharpmind-server:gpu. A local tag such as sharpmind-server:gpu cannot be pulled; setup now refuses it before trying.
- SSH fields. Host (blank uses the Base URL host), port 22, user root, key or password, container name sharpmind.
- Tick GPU host. This adds --gpus all and installs the NVIDIA container toolkit when it is missing.
- Save host.
Deploy from the API
The host card also creates and destroys the instance through the Vultr API. This is the API path: Studio makes the firewall, creates the instance, waits for the public IP and then runs the same SSH setup as above.
- Save an API key. In Settings → Providers → Compute → Vultr, paste a Vultr personal access token created with all permissions for instances and firewalls, then press Load plans. Studio checks the key before saving it; a rejected key stays in the field with the Vultr error. A saved key is stored encrypted and never shown again.
- Pick the instance. Choose a region, Ubuntu 24.04, one of the SSH keys that already exists on your Vultr account, and a plan. GPU plans start with vcg- (for example vcg-a16-6c-60g-16vram); the cheaper vc2- plans are CPU-only. GPU plans are listed first, but the panel preselects the cheapest CPU plan when it is opened from Settings, so check the plan ID before deploying. The matching private key must be selected in the host's Edit panel; Deploy stays disabled until both are present.
- Deploy instance. Studio creates a firewall with TCP 22, 5117 and 8080 open, labels the instance sharpmind-<short-id>, tags it sharpmind, and waits for it to become active. It then waits for SSH on port 22 (up to five minutes) before running the setup; a fresh instance can refuse connections for about a minute after it reports running. The first boot usually takes two to five minutes. The host card's Base URL and SSH host are filled with the new IP as soon as Vultr assigns it.
- Watch it on Compute. The left menu gains a Compute item with a count badge whenever something is running. Expand it to list every instance with its state, hourly rate and a Destroy button; a pulsing dot marks a host that is in active use (training, rehearsal, exam, or a chat/work turn running against it), and destroying an in-use instance asks for a second confirmation. The page lists the same instances with IP, uptime and estimated spend.
- Close safely. Exiting Studio warns about running instances and offers Keep running (default), Cancel and Destroy and exit. Untracked instances that carry the sharpmind tag are detected at startup and shown on Compute so they can be destroyed too.
Idle teardown
The Vultr panel's Destroy when idle choice is 15, 30 or 60 minutes, or Never. It applies per host and only to instances this Studio created. Training, rehearsal, exam, and a chat or Work turn loading or streaming from the host reset the timer. A deploy, SSH setup or model upload also counts, and the host's job list is checked every 30 seconds, so a job that outlived a Studio restart keeps the instance alive. After the idle window a banner shows a 10-minute countdown with Keep; if it expires, Studio destroys the instance and deletes its disk.
Tagged instances that no saved host tracks (another Studio install, or a host entry that was deleted) are listed as untracked. Studio never destroys them, including from Destroy all tracked and the exit prompt; remove them in the Vultr console.
Set up & deploy
On the host card press Test SSH first. It connects to the address already stored in the host card, so it only works after the instance exists and the card has its IP; before a deploy it tests whatever URL is saved. You should see the OS, Docker version, GPU name and free disk. Then press Set up & deploy. The helper streams each stage:
[connect] Connected to root@203.0.113.10:22.
[probe] Ubuntu 24.04 LTS · x86_64 · Docker 29.9.2 · NVIDIA A16 · 288G free
[pull] Pulling integral2u/sharpmind-server:gpu…
[run] Up 2 seconds | 0.0.0.0:5117->5117/tcp, 0.0.0.0:8080->8080/tcp
[ready] Container sharpmind: Up 2 seconds | integral2u/sharpmind-server:gpu
Setup replaces an existing container with the same name, so it is safe to run again after changing the image or token. The data folder is preserved.
Verify
- Press Test on the card: the dot turns on and the model list loads.
- Open http://<public-ip>:8080/health in a browser; it is unauthenticated.
- The health payload should list ilgpu with a backend such as CUDA · cuBLAS.
- If the API answers but the Studio URL does not, port 5117 is still blocked.
{
"status": "ok",
"dataRoot": "/data/SharpMind",
"accelerators": [{
"name": "ilgpu",
"inference": true,
"backend": "CUDA · cuBLAS"
}]
}
Upload and train
- Chat model. Upload a packaged .smm from the card, or leave the host without a model: a fine-tune uploads its base with the job.
- Fine-tune. Training → Improve → base .smm, corpus, SFT if the data is chat JSONL, Run on = Vultr. The export downloads back to this PC.
- Watch the engine line. GPU success looks like Engine: [Cuda] NVIDIA A16, 14962 MB, cuBLAS 12.6. If you see Retrying on the CPU, the base did not satisfy the GPU engine; see Training.
Cost and cleanup
- The A16 plan is roughly $0.47/hour; a day is about $11.
- Stopping is not enough. Destroy the instance to stop billing.
- The host's Destroy when idle window (15/30/60 minutes or Never) tears it down after a 10-minute Keep countdown. Only instances this Studio created are eligible.
- The Compute page shows the hourly rate and estimated spend per instance, and can destroy everything in one click. The left menu's expandable Compute list has a Destroy button per instance too.
- Closing Studio checks for running instances; Keep running is the default, so read the prompt before dismissing it.
- Download any exported model first. The host's /data/SharpMind is deleted with the instance unless it is a separate block volume.
- A snapshot costs a small monthly storage fee and is only worth it for data you cannot recreate.
- Rotate the host token if you shared it, and remove the instance's firewall rules with the instance.
Custom Linux host
Any Linux machine with Docker can be a compute host. The fields are the same as Vultr: Base URL, token, and optionally SSH. Use this for a workstation, a lab server, or another cloud.
- Install Docker (the SSH helper can do this) and make sure the service starts at boot.
- Open the ports 5117 and 8080, and SSH 22 if you want Studio to manage it.
- Add host → Custom remote server, enter the URL and token, and press Test.
- For GPU, install the NVIDIA driver and container toolkit; tick GPU host and let setup add --gpus all.
If the machine already runs the image, skip SSH entirely: Base URL, token, Test. The host only needs to be reachable from this PC.
Common problems
| Message | Meaning / fix |
|---|---|
| pull access denied for sharpmind-server | The image has no registry prefix. Enter user/repo:tag in Edit; setup now refuses local-only names before pulling. |
| Could not reach host:22 | The instance is down, the IP changed, or the provider firewall blocks SSH. |
| Connection refused right after the instance starts | Vultr reports active running a few seconds before SSH listens. Deploy waits for port 22 automatically; for a manual test, wait a minute and retry. |
| No NVIDIA driver found | The plan has no GPU (a vc2- plan) or the image is not a GPU image. Use a vcg- plan, or untick GPU host for a CPU-only test. |
| Invalid API token / 401 when loading plans | Vultr rejected the pasted token and Studio did not save it. Create a token with instance and firewall access, paste it again and press Load plans. |
| SSH login failed | Wrong user, wrong key, or the public key was never added to the host. The console step is the bootstrap. |
| No NVIDIA driver found | nvidia-smi is missing. Use a GPU image or install the driver and reboot; the helper does not install drivers. |
| Port 8080 or 5117 is already used | Another container owns the port. Remove it, or change the container name/ports. |
| API times out but SSH works | Provider firewall. Allow TCP 8080 and 5117, not just 22. |
| 401 after rotating a token | Restart Studio so it reloads the encrypted token, then Test. The old token is invalid by design. |