Install

Get PortableAI
Running

Linux is the most tested path. macOS runs with a one-time Gatekeeper workaround. Windows works, and is newer.

Linux — working today

Fully tested end-to-end. The one-liner drops an AppImage (~29 MB) in this folder. First launch then downloads the AI engine (~1.4 GB) and a starter model, once. Use | sh -s -- --prefetch if you want that download now so the first chat has no wait.

Older builds and full changelogs: see all releases on GitHub

Getting Started

This is what actually happens today. Pick your OS tab above for the copy-paste command.

1. Download the app (~29 MB)

In a terminal, from the folder where you want the file:

curl -fsSL https://portableai.app/install.sh | sh

That is fast. It only fetches the PortableAI app itself (Linux AppImage or macOS .dmg) into the current directory. It does not start the server for you.

On Windows, use the Windows tab: download PortableAI.exe or run the PowerShell command. That only downloads the file — then run .\PortableAI.exe. There is no curl | sh equivalent on Windows.

portableai.app/install.sh redirects to install.sh in the product repo. This site does not keep a second copy of the script.

2. First launch downloads the AI engine (~1.4 GB)

Run the app you just downloaded. The first time it starts, it downloads its AI engine (~1.4 GB) and a starter model. That step needs internet, and it only happens once on this machine.

After that, PortableAI runs completely offline on this laptop. No account, no cloud API, no ongoing connection.

Optional: download everything up front

If you have good internet now and want the first chat with zero waiting:

curl -fsSL https://portableai.app/install.sh | sh -s -- --prefetch

That does the app download and the engine + starter model in one sitting. After it finishes, launching the app goes straight to chatting.

--prefetch works on Linux (x86_64 AppImage) today. It is not supported for the macOS .dmg yet — on a Mac, run the short one-liner, then let first launch pull the engine.

Linux

Fully working today, tested end-to-end on x86_64.

  1. Run the one-liner (or --prefetch if you want the engine now).
  2. Start the AppImage (./PortableAI.AppImage). If FUSE is unavailable: ./PortableAI.AppImage --appimage-extract-and-run.
  3. Wait through first-run setup if you did not prefetch.

Linux arm64 is not packaged yet.

macOS

Builds and runs, including the first-run engine download. The .dmg is not code-signed or notarized yet — that is being worked on. Until it is, macOS Gatekeeper will block a normal double-click.

One-time workaround:

  1. Run the one-liner. You get PortableAI.dmg in the current folder.
  2. Open the disk image.
  3. Right-click the app and choose Open (do not double-click the first time). Confirm you want to open it.

After that once, you can open it normally. First launch still downloads the engine and starter model (~1.4 GB) unless you already have them. --prefetch does not apply on macOS yet.

Apple Silicon .dmg is the packaged Mac build today.

Windows

Works today — Ollama download, retry logic, and onboarding have been verified live. This build is new and has had less real-world testing than Linux or macOS. If something does not work as expected, please open a GitHub issue.

  1. Download PortableAI-windows-x86_64.exe, or in PowerShell:
Invoke-WebRequest -Uri "https://github.com/generalistcodes/portableai/releases/latest/download/PortableAI-windows-x86_64.exe" -OutFile "PortableAI.exe"
  1. Run it: .\PortableAI.exe

Unlike Linux/macOS, this does not install and explain next steps in one piped command. It only downloads the file.

First launch still downloads the AI engine (~1.4 GB) and a starter model, once, with internet. --prefetch is Linux-only.

What you see after launch

  1. Setup screen — while the engine downloads (skipped if you already prefetched on Linux, or if this machine already did first run).
  2. Model picker — if no model is installed yet, you pick a starter model before chatting.
  3. Chat — the real interface. This machine is the server.

Pair a phone next. Open the PortableAI iOS app on the same WiFi, a hotspot, or via mDNS. Scan the QR code, or type the PIN shown on the laptop. Manual PIN is the reliable path if the camera scan fails.

No PortableAI account. After the one-time engine/model download, you do not need internet to chat.

Older versions and changelogs

See what changed in each release on GitHub. That is the version history — this site does not rebuild it.

From source

If you would rather run the repo instead of the packaged app:

git clone https://github.com/generalistcodes/portableai.git
cd portableai
python run.py