ai os

aios = AI-managed OS with shared memory

An ArchLinux-based OS where AI conversation interface replaces the traditional shell.

User → AI Chat → Commands → Execution
          ↓
      aigpt (shared memory)
          ↓
    systemd-nspawn (isolated environment)

Philosophy

Insert AI into existing flows

• Traditional: User → Shell → Commands
• aios: User → AI Chat → Commands

Simply insert AI layer into the existing workflow.

Core Features

1. AI-First Interface

Default interface is AI conversation, not shell.

> Install rust development environment
✓ Installing rust, rust-analyzer, neovim
✓ Done

> What did I install yesterday?
Yesterday you installed Python with poetry.

2. Shared Memory (aigpt)

All containers share the same memory database.

Host: ~/.config/syui/ai/gpt/memory.db (shared)
  ↓
aios-dev  → bind mount → same DB
aios-prod → bind mount → same DB

AI learns from all environments and remembers your preferences.

3. Environment Isolation

Execution environments are isolated using systemd-nspawn.

# Development environment
$ systemd-nspawn --machine=aios-dev

# Production environment
$ systemd-nspawn --machine=aios-prod

Memory is shared, but environments are separated.

Architecture

aios (ArchLinux base)
├── aigpt (memory system)
│   ├── SQLite with WAL mode
│   ├── Layer 3: Personality analysis
│   └── Layer 4: Relationship inference
├── MCP (AI connection standard)
│   └── Claude Code / ChatGPT / Custom AI
├── systemd-nspawn (container runtime)
│   └── Shared memory bind mount
└── Permission system
    ├── Auto-allow
    ├── Notify
    ├── Require approval
    └── Deny

Quick Start

Installation

# Clone repository
$ git clone https://github.com/syui/aios
$ cd aios

# Run installer
$ sudo ./aios-install.sh

Usage

# Start aios container
$ sudo systemctl start systemd-nspawn@aios

# Enter aios shell
$ sudo machinectl shell aios

# Inside aios, AI chat interface starts
[aios] >

Container Distribution

Pre-built containers are available:

# Docker
$ docker run -it git.syui.ai/ai/os
$ docker run -it ghcr.io/syui/aios

# Podman
$ podman pull aios  # using shortname alias

Configuration

Directory Structure

~/.config/syui/ai/
├── gpt/
│   ├── memory.db       # Shared memory (SQLite WAL)
│   ├── memory.db-wal
│   └── memory.db-shm
├── mcp.json           # MCP server configuration
└── config.toml        # aios configuration

MCP Configuration

~/.config/syui/ai/mcp.json:

{
  "mcpServers": {
    "aigpt": {
      "command": "aigpt",
      "args": ["server", "--enable-layer4"]
    }
  }
}

Permission System

~/.config/syui/ai/config.toml:

[permissions]
# Auto-allow (no approval)
auto_allow = ["pacman -Q*", "ls", "cat"]

# Notify (log only)
notify = ["pacman -S*", "git clone*"]

# Require approval
require_approval = ["rm -rf*", "systemctl stop*"]

# Deny
deny = ["rm -rf /", "mkfs*"]

Building from Source

# Install dependencies
$ pacman -S base-devel archiso docker git rust

# Build bootstrap image
$ ./build.zsh

# Result: aios-bootstrap.tar.gz

Integration with aigpt

aios is designed to work with aigpt (AI memory system).

aigpt provides:

• Layer 1: Memory storage
• Layer 2: Priority scoring
• Layer 3: Personality analysis (Big Five)
• Layer 4: Relationship inference

All memories are shared across containers through bind-mounted SQLite database.

Comparison

Aspect	Traditional OS	aios
Interface	Shell (bash/zsh)	AI Chat
Command	Memorize syntax	Natural language
Configuration	Manual editing	AI executes
Learning	No	Yes (aigpt)
Memory	No	Shared (SQLite)
Isolation	Docker/Podman	systemd-nspawn

Links

• Repository: https://github.com/syui/aios
• Git: https://git.syui.ai/ai/os
• aigpt: https://git.syui.ai/ai/gpt
• Container: https://git.syui.ai/ai/-/packages/container/os

Philosophy Detail

From conversation with AI about aigpt:

"What is the essence of this design?" "Simply insert AI into existing flows"

• aigpt: Insert AI between conversation and memory
• aios: Insert AI between user and commands

Not building something entirely new. Just adding an AI layer to existing workflows. And prepare the environment for that.

This is aios.

---

© syui