English · 简体中文 · User Guide

A zero-dependency operating system that turns capable AI Code Agents into reliable software delivery systems. Install it once across Claude Code, Codex, GitHub Copilot, Cursor, and Qoder—then give every Agent the same project memory, delivery workflow, human approval gates, and verification discipline.

Prerequisites: Git and Bash on macOS/Linux, or Git and PowerShell on Windows. AI Coding OS does not require Node/npm, Python, a server, or a database.

macOS / Linux (Bash):

git clone https://github.com/DOS-AI-Tech/AI-Coding-OS.git ~/ai-coding-os
bash ~/ai-coding-os/installer/install.sh

Windows (PowerShell):

git clone https://github.com/DOS-AI-Tech/AI-Coding-OS.git "$HOME\ai-coding-os"
& "$HOME\ai-coding-os\installer\install.ps1"

Choose the Agents you use. The installer adds the core framework and five basic skills to their global skill directories for the current user.

The target project directory must already exist.

macOS / Linux (Bash):

bash ~/ai-coding-os/installer/init.sh /path/to/project

Windows (PowerShell):

& "$HOME\ai-coding-os\installer\init.ps1" "C:\path\to\project"

This creates only project-owned memory, domain scaffolding, output folders, and thin Agent adapters. Global skills are not copied into the project.

If ~/ai-coding-os (or $HOME\ai-coding-os on Windows) already exists, update it instead of cloning it again.

macOS / Linux (Bash):

cd ~/ai-coding-os
git pull
bash installer/update.sh

Windows (PowerShell):

Set-Location "$HOME\ai-coding-os"
git pull
& ".\installer\update.ps1"

Open the project with any selected Agent and describe a task. AI Coding OS reads the project state, chooses the right workflow, presents the required human gate, and only then proceeds.

AI Code Agents can already analyze requirements, design architecture, plan work, write code, run tests, and deploy software. The unresolved problem is not whether an Agent can perform each task—it is whether the Agent can execute the whole lifecycle consistently, in the right order, with enough context and evidence to trust the result.

Without operating discipline, capable Agents still fail in predictable ways:

The goal is simple: let Agents do the work while humans retain control—without making the human repeatedly supply memory, enforce the process, and discover omissions at the end.

AI Coding OS includes shared workflow routing, memory conventions, output templates, project initialization, and five built-in delivery capabilities. They let the same Agent adopt the right mode for each stage instead of improvising the entire project from a single instruction:

Installed once for each selected Agent
└── AI Coding OS core + built-in skills
                      │
                      ▼
Initialized once per project
├── .ai/memory/       shared project state
├── domain/           project-specific knowledge
├── project_docs/     requirements, plans, and test evidence
└── thin adapters     ensure each Agent activates the global framework

Static framework behavior is global. Business context and work history stay inside the project, where teams can review and version them with Git.

Advanced Skill Packs extend AI Coding OS with specialized capabilities for domain discovery, business modeling, technical analysis, solution design, and expert review workflows.

Download: —

• English User Guide
• 中文使用指南
• Skill format reference
• Workflow reference

Updates leave project memory untouched.

Claude Code, OpenAI Codex, GitHub Copilot, Cursor, and Qoder are supported through the shared Agent Skills format and native global directories defined in installer/agents.conf.

AI Coding OS is released under the MIT License. Issues and pull requests are welcome.