Skip to content

PAI Quickstart

Goal: a running PAI in under ten minutes. For background and concepts, read getting-started.md instead.

You will need:

  • A USB stick, 16 GB or larger (32 GB recommended for persistence).
  • A computer with a USB port and UEFI firmware.
  • About ten minutes.

1. Check hardware requirements

  • 8 GB RAM minimum, 16 GB recommended for local LLMs.
  • A free USB-A or USB-C port (USB 3.x strongly recommended).
  • UEFI firmware (almost every machine since ~2015). Legacy BIOS-only machines are not supported.

2. Download the ISO and signature

Grab the latest pai-<version>.iso and its .minisig signature from the project's release page. Put both files in the same directory.

3. Verify the download

Check the SHA-256 first, then the minisign signature:

sha256sum pai-<version>.iso
minisign -Vm pai-<version>.iso -P <release-signing-pubkey>

Both must succeed. If either fails, stop and re-download. See ../SECURITY.md for the signing key.

4. Flash the ISO

Pick the tool that matches your OS:

  • Linux / macOS — sudo dd if=pai-<version>.iso of=/dev/sdX bs=4M status=progress conv=fsync. Replace /dev/sdX with your stick. The wrong device will destroy its contents; check with lsblk first.
  • Any OS — BalenaEtcher. Pick ISO, pick stick, click Flash.
  • Windows — Rufus. Select ISO, leave defaults, write in DD mode.

See usb-flashing.md for screenshots and per-OS detail.

5. Boot from USB

Restart with the stick inserted. Press the boot-menu key for your vendor:

Vendor Key
Dell F12
HP F9 / Esc
Lenovo (ThinkPad) F12
ASUS F8 / Esc
Acer F12
MSI F11
Apple (Intel) hold Option

Pick the USB entry. If your machine boots to the host OS instead, see troubleshooting.md#boot-issues.

6. First-boot setup

The live system asks for:

  • Keyboard layout.
  • Whether to enable persistence (recommended — an encrypted partition on the stick for chats, models, wallet keys). Set a strong passphrase. There is no recovery if you forget it.

A minute later you'll land at the Sway desktop.

7. Hello, AI

Open a terminal with Super + Enter and run:

ollama run llama3.2

First load takes a few seconds. Then type a prompt, hit enter, and you're talking to a model running entirely on your hardware. Ctrl+D exits.

You're done

From here: