AI-2 – Artix-based Linux distribution with local AI assistant

AI-2 is an Artix Linux-based distribution designed to run local large language models on ordinary and older PCs. It uses the Xfce desktop and runit init system, and combines a lightweight Linux environment with tools that detect the computer’s capabilities and select an appropriate AI configuration.

The distribution focuses on CPU-based inference rather than requiring a powerful discrete GPU. Its setup wizard analyses the processor, memory and other hardware, assigns one of six capability tiers, applies system tuning, installs the appropriate llama.cpp build and recommends models suited to the machine. An AI Score derived from benchmarking provides an additional measure of what the computer can realistically run.

Key Features

• Runs language models locally using llama.cpp.
• Supports CPU-based inference, including older processors without AVX or SSE4.1.
• Provides separate llama.cpp builds for different CPU instruction sets.
• Detects CPU, RAM, GPU, storage and init-system information.
• Assigns hardware to Tiny, Light, Standard, Creator, Studio or Workstation capability tiers.
• Benchmarks inference performance and generates an AI Score.
• Recommends models according to measured hardware performance.
• Setup wizard can configure zram, earlyoom and other system settings.
• Downloads, verifies, lists and removes locally installed models.
• Browser-based interface for chatting with local models.
• Provides an OpenAI-compatible local API through llama-server.
• Live environment with graphical installation supporting BIOS and UEFI systems.
• 64-bit x86 PCs are supported with 2 GB RAM minimum and 4 GB recommended.
• Uses pacman for package management and follows the Artix rolling-release model.