System requirements

Minimum requirements

• CPU: 2 cores, 64-bit (x86_64/amd64 or arm64)
• RAM: 4 GB (8 GB recommended; 16 GB+ for local AI models)
• Storage: 20 GB free disk space (more for media and model weights)
• Architecture: amd64 or arm64

Platform support

Docker backends

Hub does not require Docker Desktop. Docker Desktop needs a paid subscription for organizations with more than 250 employees or more than $10M in annual revenue, so every platform also has a licensing-free option:

Hub can install the alternative for you — the desktop app’s Docker setup screen offers Auto-Install Docker Engine in WSL2 on Windows and Auto-Install Colima on macOS.

Pick a backend by platform and licensing:

flowchart LR
    P{"Platform?"}
    P -->|"Windows"| W{"Licensing-free?"}
    P -->|"macOS"| M{"Licensing-free?"}
    P -->|"Linux"| L["Docker Engine<br/>+ Compose plugin"]
    W -->|"no"| WD["Docker Desktop<br/>WSL2 or Hyper-V backend"]
    W -->|"yes"| WE["Docker Engine in WSL2<br/>Hub can install it"]
    M -->|"no"| MD["Docker Desktop"]
    M -->|"yes"| MC["Colima<br/>Hub can install it"]

Windows

Either backend works, and WSL2 is usually involved either way: Docker Desktop defaults to its WSL2 backend, and Hub installs it that way. Docker Desktop’s Hyper-V backend (Windows Pro and Enterprise) is also supported.

• Docker Desktop — install it, start it, and Hub detects it automatically.
• Docker Engine in WSL2 — Hub provisions an Ubuntu WSL2 distro, installs the open-source Docker Engine inside it, exposes the daemon on 127.0.0.1:2375, and selects it through a wsl-engine Docker context. No Docker Desktop license required.

If WSL has to be enabled during setup, restart Windows before reopening Hub.

Linux server distributions

CI-OS also provisions Linux Mint, Fedora, Arch, Bazzite, and NixOS. Ubuntu is the reference implementation. Bazzite ships as an image and NixOS as a set of Nix modules, rather than as install scripts.

Software requirements

• Docker — Docker Desktop  or Docker Engine  (see Docker backends for the licensing-free options on each platform). Install Docker is the step-by-step walkthrough.
• Docker Compose plugin, version 2.33 or newer — Verify with docker compose version. On an older Compose, Hub refuses to start and names the version it found. Docker Engine 28 or newer is recommended: on an older Engine, Hub starts but warns that its containers may use the wrong network for internet traffic.
• A local inference engine — Ollama  is the default, and the engine vLLM and oMLX borrow for embeddings. Hub also supports vLLM on NVIDIA GPUs, oMLX on Apple silicon, and Lemonade on AMD GPUs. You need one for local AI models; cloud provider keys need none. See AI and inference

Optional:

• NVIDIA drivers + Container Toolkit — For CUDA GPU acceleration
• ROCm — For AMD GPU acceleration on Linux

Network requirements

• Outbound internet for pulling Docker images and App Store catalog
• Port 5002 on the host for the Hub dashboard and API
• Port 80/443 on the host for Traefik, the reverse proxy for exposed apps. Traefik uses them when they’re free and moves to 8880 and 8443 when they aren’t. The Linux desktop app always uses 8880 and 8443, because Linux doesn’t let a normal user open ports below 1024
• No inbound ports required for CI Account pairing or Tailscale

Not supported

• 32-bit systems
• Windows containers — Hub runs Linux containers only
• Bare-metal Windows or macOS as a server host — use the desktop app instead

Companion hardware

Companion Intelligence sells purpose-built hardware preloaded with Hub. See ci.computer  for current devices and Core Server quick start for headless setup.

What you supply on each route: