Run a Jupyter Notebook / JupyterLab Server on a VPS Securely

JupyterLab (the modern successor to Jupyter Notebook) is the standard interactive environment for Python data analysis, machine learning experiments, teaching and quick scripting. Running it on a VPS gives you an always-on notebook server: long jobs keep running when your laptop sleeps, data stays next to the compute, and you can reach it from any browser.

Requirements

1–2 vCPU and 2 GB RAM for light analysis. Data work is often RAM-bound: pandas typically needs several times the size of the dataset in memory. See how much RAM a VPS needs.
For deep learning training you want a GPU. A CPU VPS is fine for data prep, small models and inference.

Install in a virtual environment

sudo apt install -y python3-venv python3-pip
python3 -m venv ~/jupyter
. ~/jupyter/bin/activate
pip install jupyterlab pandas numpy matplotlib

Set a password (stored hashed in ~/.jupyter/jupyter_server_config.json):

jupyter lab password

See Python virtual environments.

Option 1: SSH tunnel (simplest and safest)

jupyter lab --no-browser --ip 127.0.0.1 --port 8888
# on your computer:
ssh -L 8888:127.0.0.1:8888 user@server
# then open http://localhost:8888

Nothing is exposed to the internet. See SSH tunnelling.

Run it as a service

# /etc/systemd/system/jupyter.service
[Unit]
Description=JupyterLab
After=network.target

[Service]
User=alice
WorkingDirectory=/home/alice/notebooks
ExecStart=/home/alice/jupyter/bin/jupyter lab --no-browser --ip 127.0.0.1 --port 8888
Restart=on-failure

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload && sudo systemctl enable --now jupyter

Kernels keep running across browser disconnects. That’s the point of a server. See systemd services.

Option 2: Nginx with HTTPS

If you need browser access without a tunnel:

server {
    server_name jupyter.example.com;
    client_max_body_size 200M;
    location / {
        proxy_pass http://127.0.0.1:8888;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host $host;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_read_timeout 86400;
    }
    # listen 443 ssl; certbot --nginx -d jupyter.example.com
}

WebSockets are required: kernels talk over them. Add to ~/.jupyter/jupyter_server_config.py:

c.ServerApp.trust_xheaders = True
c.ServerApp.allow_remote_access = True

Security warning: a Jupyter session is arbitrary code execution as that user. Use a strong password, HTTPS only, and ideally a VPN or SSO in front. Scanners look for open Jupyter servers to install crypto miners. See Tailscale and Nginx basic auth.

Kernels and environments

python3 -m venv ~/envs/ml && ~/envs/ml/bin/pip install ipykernel scikit-learn
~/envs/ml/bin/python -m ipykernel install --user --name ml --display-name "Python (ml)"

Kernels exist for R (IRkernel), Julia, and others too.

JupyterHub for teams

For several users, use JupyterHub: each user gets their own server and login (Linux users, OAuth/GitHub, or LDAP). The Littlest JupyterHub (TLJH) sets this up on a single VPS in one command, with HTTPS and per-user memory/CPU limits. Use it for classrooms and small data teams.

Resource management

Long jobs: run them in the background, and save results to disk. Notebooks aren’t a job scheduler, so convert heavy jobs into scripts with jupyter nbconvert --to script, and run them with cron or systemd.
Watch memory: an out-of-memory kernel dies silently. Add swap as a safety net. See swap and htop.
Shut down idle kernels: c.MappingKernelManager.cull_idle_timeout = 3600.

Backups

Notebooks are files: keep them in Git (with outputs stripped, using nbstripout) and back up the data directory. See backup strategy.

Quick reference

Task	Command
Install	`pip install jupyterlab` (in a venv)
Set password	`jupyter lab password`
Run locally bound	`jupyter lab --no-browser --ip 127.0.0.1`
Tunnel	`ssh -L 8888:127.0.0.1:8888 user@server`
Add a kernel	`python -m ipykernel install --user --name X`

Development
Python
Self-Hosting