PAI

What PAI is

PAI is a bootable live USB operating system built on Debian 12 that gives you a private, portable AI workstation. Plug it into any x86-64 PC, boot from USB, and you have a complete AI environment that leaves no trace behind.

Like Tails is to privacy-focused browsing, PAI is to private AI. It combines the amnesic, leave-no-trace philosophy of Tails with the power of local large language models running via Ollama.

Your conversations exist only in RAM while PAI is running. Shut down, and they're gone — no logs, no history, no server-side copies. The machine doesn't even know PAI was there.

What you get

• 🤖
  Offline AI assistant — Ollama runs Llama, Mistral, Phi, Gemma, and 100+ models locally. No API keys, no cloud, no data leaving your machine.
• 🔒
  Privacy-hardened environment — MAC spoofing on every boot, UFW firewall denying all inbound connections, optional Tor routing.
• 🖥️
  Complete desktop — Sway (Wayland tiling WM), Firefox ESR, terminal (foot), file manager, text editor. Everything you need, nothing you don't.
• 💾
  Zero installation — boots entirely from USB into RAM. Nothing touches the host machine's hard drive.
• 🔑
  Crypto wallets — Electrum (Bitcoin) and Monero wallet pre-configured with sane privacy defaults.
• 🛡️
  Encrypted persistence (opt-in) — save AI models, SSH keys, and chat history to a LUKS2-encrypted partition on the USB across sessions.

Who it's for

• Privacy-conscious users who want to use AI without sending prompts to corporate servers.
• Journalists and researchers who work with sensitive sources and documents.
• Security professionals who need an isolated, trusted environment for sensitive analysis.
• Crypto users who want to manage wallets on a network-isolated machine.
• Travelers who want to use AI on any computer without leaving traces.
• Developers who want to experiment with local LLMs without cloud dependencies or costs.

What it's not

PAI raises the cost of surveillance significantly, but informed expectations matter:

• Not a GPU rig — CPU-only inference. Expect 5–20 tokens/sec for 7B models. Fast enough for most tasks; use smaller models for speed.
• Not Tails — Tails routes everything through Tor by default. PAI makes Tor opt-in. PAI's focus is private AI, not anonymous browsing — though you can combine both.
• Not a permanent install — PAI is a live system. It boots from USB. No traditional hard drive installation.
• Not protection against physical attacks — hardware keyloggers, compromised BIOS/UEFI firmware, and cold boot attacks are out of scope.
• Not cloud AI — you get what runs on your hardware. Local models are powerful and improving rapidly, but they're not GPT-4.