Get the latest setups
Start Here/Guides
SOFTWARE GUIDE · UPDATED OCTOBER 2026

Run open-source AI apps with Pinokio

Pinokio is a free desktop app that installs and launches open-source AI apps in one click: image and video studios, voice tools, chat apps, and more. Run it on your own computer, or on a rented GPU when your machine is not powerful enough.

Choose where to run it No GPU? Use RunPod
WHAT YOU WILL HAVE
  • Pinokio installed and set up
  • An app installed from Discover in one click
  • That app started and open in Pinokio
  • Apps saved where they survive restarts
01 · BEFORE YOU BEGIN

What your machine needs

Pinokio itself is light. The requirements come from the apps you install, and every app’s listing states its platform and GPU needs. Check the app first, then pick a route.

RouteGood forKeep in mind
RunPod (cloud GPU)Any computer with a browser. Runs NVIDIA-only apps such as Maestro (listed at 6 GB+ VRAM).Billed hourly while running. Your files live on the rented machine.
Windows PCNVIDIA GPUs; AMD where an app offers an AMD version.Large apps can take tens of GB of disk each.
MacApps marked for Apple / macOS.NVIDIA-only apps will not run.
LinuxNVIDIA or AMD on Ubuntu 22.04+, Debian 12+, Fedora.Needs glibc 2.34+; pick the correct x64 or ARM64 package.

Browse apps and their requirements in the Pinokio app catalog , which filters by platform (macOS, Windows, Linux) and GPU (NVIDIA, AMD, Apple).

02 · CHOOSE WHERE IT RUNS

Install Pinokio

Pick one route. RunPod is the way to use apps that need more GPU than your computer has.

Run Pinokio on a rented GPU, in your browser

You rent an NVIDIA GPU by the hour, open a full Linux desktop in a browser tab, and use Pinokio exactly as you would at home. Your own computer only needs a modern browser. Apps such as Maestro that require an NVIDIA GPU run on the rented card.

Files and prompts you use in Pinokio live on the rented machine, not on your computer. RunPod bills while the pod runs; storage is billed while it is stopped.

OPTION A · EASIESTDeploy the YouRunAI Pinokio Desktop

A ready-made image with the desktop and Pinokio preinstalled. Pinokio opens by itself after you log in.

  1. 1

    Create a pod from the image

    Open RunPod

    Go to Pods → Deploy, then Change Template → Edit Template (or create a custom template) and set:

    Container imageghcr.io/lucid-digital/pinokio-desktop:latest
    Container start commandLeave empty
    Expose HTTP ports6080
    Volume150 GB mounted at /workspace
    Environment variableVNC_PASSWORD = a password of up to 8 characters
  2. 2

    Pick a GPU and deploy

    Filter the GPU list by the app you plan to run. Maestro, for example, lists an NVIDIA GPU with 6 GB+ of VRAM. A 24 GB card such as an RTX 4090 or RTX 3090 leaves room for most Pinokio apps.

    Use RunPod’s CUDA version filter to pick a machine with CUDA 12.8 or newer (13.x for RTX 50-series cards). Maestro’s environment uses PyTorch built for CUDA 12.8, and older hosts fail with a “driver too old” error. Maestro also asks for at least 150 GB of free disk (500 GB recommended), so size the volume for the apps you plan to install.

  3. 3

    Open the desktop

    When the pod is running, click Connect → HTTP Service [Port 6080]. Enter your VNC_PASSWORD. If you left it empty, a generated password is printed in the pod's Logs tab.

    A RunPod “Waiting for service to respond” page for the first minute or two is normal while the image downloads and the desktop starts.

  4. 4

    Let Pinokio open

    Pinokio starts automatically on the desktop. Continue with First run and set its home folder to /workspace/pinokio.

OPTION B · MANUALAdd a desktop to RunPod’s official PyTorch template

Use this if you prefer RunPod’s own template. It takes about five minutes and one paste. YouRunAI ran these steps on a RunPod PyTorch pod (Ubuntu 24.04) on October 20, 2026 and reached a working desktop with Pinokio.

  1. 1

    Deploy the PyTorch template

    In RunPod, choose a GPU and the official Runpod PyTorch template with an Ubuntu 22.04 or 24.04 image. Click Edit Template, add 6080 to Expose HTTP Ports, and set the volume to 100 GB or more. Deploy.

  2. 2

    Open a terminal in the pod

    Click Connect → Jupyter Notebook, then New → Terminal. Processes started there keep running after you close the tab.

  3. 3

    Paste the setup

    Pod terminal
    export DEBIAN_FRONTEND=noninteractive
apt-get update && apt-get install -y xfce4 xfce4-terminal dbus-x11 tigervnc-standalone-server tigervnc-tools novnc websockify wget curl pciutils dmidecode
URL=$(curl -s https://api.github.com/repos/pinokiocomputer/pinokio/releases/latest | grep browser_download_url | grep amd64.deb | cut -d '"' -f 4)
wget -O /tmp/pinokio.deb "$URL" && apt-get install -y /tmp/pinokio.deb
mkdir -p ~/.vnc /workspace/pinokio
printf '#!/bin/sh\nunset SESSION_MANAGER DBUS_SESSION_BUS_ADDRESS\nexec dbus-launch startxfce4\n' > ~/.vnc/xstartup && chmod +x ~/.vnc/xstartup
vncserver :1 -geometry 1920x1080 -localhost yes -xstartup ~/.vnc/xstartup
nohup websockify --web=/usr/share/novnc 6080 localhost:5901 >/tmp/novnc.log 2>&1 &
sleep 3; echo "--- CHECK ---"; which pinokio; curl -sI localhost:6080/vnc.html | head -1

    When asked, type a desktop password twice (up to 8 characters) and answer n to the view-only password. The check at the end should print /usr/bin/pinokio and HTTP/1.1 200 OK.

  4. 4

    Open the desktop and start Pinokio

    Click Connect → HTTP Service [Port 6080], add vnc.html to the end of the address, click Connect, and enter your password. In the desktop, open Applications → Terminal Emulator and run:

    Desktop terminal
    pinokio --no-sandbox

    The desktop software is installed on the pod’s container disk, so a pod restart removes it. Run the setup again after a restart; Pinokio apps in /workspace/pinokio remain.

Stop, don’t terminate. When you finish, click Stop on the pod. Your volume keeps Pinokio and its apps for a small storage fee. Terminate deletes everything.

03 · FIRST RUN

Set up Pinokio and install your first app

  1. 1

    Choose Pinokio’s home folder

    On first launch Pinokio asks where to keep its files. Every app you install, and its models, go here, so choose a location with plenty of free space. On RunPod, use /workspace/pinokio: that is the volume that survives stopping the pod.

  2. 2

    Let it finish setting up

    Pinokio downloads its own tools, such as Python, Node.js, Git, Conda, and FFmpeg, into that folder. Leave it running until it finishes. Nothing needs to be installed separately.

  3. 3

    Find an app in Discover

    Open Discover and search for the app you want, for example Maestro, an all-in-one local AI creative studio. Check the requirements on its listing.

  4. 4

    Install, then Start

    Click Install. Pinokio downloads the app and its dependencies; large apps can take from a few minutes to more than half an hour and use tens of gigabytes. When it finishes, click Start.

04 · CHECK IT WORKS

You are done when the app’s interface opens and produces a result

  • After Start, the app’s own interface opens inside Pinokio. Create one output (an image, a clip, a reply) and confirm it appears.
  • Using it again: open Pinokio, click the app under your installed apps, and click Start. Wait for the app’s Open Web UI button before opening it; opening it early shows “Disconnected”. Use the app’s Update item to get new versions.
  • Shutting down: stop the app in Pinokio, then close Pinokio. On RunPod, also click Stop on the pod.
  • Where things are saved: everything lives under Pinokio’s home folder, with each app in its own folder under api.
05 · TROUBLESHOOTING

Fixes for common problems

GLIBC_2.34 not found when Pinokio starts

The operating system is too old for current Pinokio. Use Ubuntu 22.04 or newer. On RunPod, the older “RunPod Desktop” (Kasm) templates run Ubuntu 20.04 and hit this error; use Option A or B above instead.

“Unmet dependencies … not installable” when installing the .deb

You downloaded the ARM64 package for an x64 machine, or the reverse. Run uname -m and download the matching file (amd64 for x86_64). RunPod GPU machines are x64.

RunPod shows “Waiting for service to respond” or a 404 on port 6080

Nothing is serving the desktop yet. Give a new pod a few minutes. For the manual route, check that the setup check printed HTTP/1.1 200 OK. After a pod restart, start the desktop again:

Pod terminal
vncserver -kill :1 >/dev/null 2>&1; pkill -f websockify
vncserver :1 -geometry 1920x1080 -localhost yes -xstartup ~/.vnc/xstartup
nohup websockify --web=/usr/share/novnc 6080 localhost:5901 >/tmp/novnc.log 2>&1 &
Can't exec /root/.vnc/xstartup

The desktop start file was not created. Run the mkdir and printf lines from the setup again, then the restart commands above.

Pinokio does not open when started from a terminal as root

Electron apps refuse to run as root without a flag. Start it with pinokio --no-sandbox.

An install stops partway or the disk fills up

AI apps and their models are large. Free space on the drive with Pinokio’s home folder, or on RunPod stop the pod, increase the volume under Edit Pod, and start it again. Then use the app’s Install or Reinstall item.

On RunPod or Linux, an app says it needs an NVIDIA GPU (or shows “Disconnected”) even though you have one

Pinokio could not detect the GPU, usually because the hardware-detection tools are missing. This happened on a RunPod A100 during YouRunAI’s setup, and installing them fixed it. First confirm the GPU and note the CUDA version (top right; 12.8 or newer for Maestro):

Terminal
nvidia-smi

Install the detection tools and check that the GPU is listed:

Terminal (add sudo if not root)
apt-get update && apt-get install -y pciutils dmidecode
lspci | grep -i nvidia

Then restart Pinokio completely, because it checks hardware only at startup. Close it (or run pkill -f pinokio), start it again from the desktop with pinokio --no-sandbox (or the desktop icon), open the app, click Update, then Start, and wait for Open Web UI. The YouRunAI Pinokio Desktop image already includes these tools.

An app says it needs CUDA or an NVIDIA GPU on a Mac or AMD machine

That app does not support your hardware. Check its listing for an AMD or Apple version, or run it on RunPod with an NVIDIA GPU.

06 · SOURCES AND FILES

Official links

RunPod links on this page are affiliate links. Hardware notes come from each app’s Pinokio listing; YouRunAI has not measured app speeds.