🚀 Ultra-Simple Setup - One command, choose your IDE, and you're ready to code with AI!

Share one AI configuration across all your AI coding tools - write once, use everywhere

# 1. Install agnostic-ai
curl -fsSL https://raw.githubusercontent.com/betagouv/agnostic-ai/main/install.sh | bash

# 2. Migrate your existing documentation (optional)
.ai/cli migrate             # Shows instructions

# 3. Use commands to build features
# In your IDE (Claude Code/Cursor):
/core:feature-create        # Scaffold new features with EPCT methodology
/core:command-create        # Create custom commands
/core:agent-create          # Create specialized agents

# 4. Enable MCP servers (optional - AI superpowers!)
.ai/cli mcp install         # Install MCP templates
.ai/cli mcp list            # See available servers (GitHub, Notion, Airtable, etc.)
.ai/cli mcp use github      # Enable GitHub MCP server
.ai/cli mcp unuse github    # Disable when not needed (saves tokens!)

That's it! Your AI configuration is ready and version-controlled in .ai/

Modern development teams face a configuration nightmare when using AI assistants:

Different IDEs, Different Configs 🤯

• Your team uses Claude Code, Cursor, Windsurf, or other AI tools
• Each tool requires its own configuration format
• You end up duplicating commands, agents, and guidelines across tools
• Changes need to be manually synced everywhere
• No single source of truth = configuration drift

AI CLI provides a single .ai/ folder that works with all AI coding tools, with a modular plugin system to install only what you need.

• ✅ Ultra-fast setup - One question: which IDE?
• ✅ Modular plugins - Install only what you need
• ✅ Write once, use everywhere - One configuration, all IDEs
• ✅ Git-friendly - Commit only .ai/, configs are generated
• ✅ Easy updates - .ai/cli update to get latest
• ✅ Team synchronization - Everyone gets the same setup
• ✅ Works everywhere - Any bash system (macOS, Linux, WSL)

curl -fsSL https://raw.githubusercontent.com/betagouv/agnostic-ai/main/install.sh | bash

What happens:

1. Creates .ai/ structure

   • Sets up AGENTS.md (main configuration)
   • Creates context/, commands/, agents/, avatars/ folders

2. Installs core plugin automatically

   • Essential commands: /core:migrate, /core:command-create, /core:agent-create, etc.
   • Essential agents: explore-codebase, prompt-engineering, fast-coder

3. Creates .ai/config.jsonc (committed to git)

   • Stores installed plugins
   • Shared across the team with JSONC format (supports comments)

4. Asks which IDE(s) you want to configure

   • Claude Code
   • Cursor
   • You can select multiple (e.g., "1 2" for both)
   • Creates symlinks for each selected IDE
   • Updates .gitignore automatically

your-project/
├── .ai/                          # ✅ Commit this (source of truth)
│   ├── AGENTS.md                 # Main config file
│   ├── config.jsonc              # ✅ Configuration (committed, supports comments)
│   ├── cli                       # Plugin manager CLI
│   ├── commands/                 # Commands (plugins + custom)
│   │   ├── core/                 # Core plugin
│   │   ├── git/                  # Git plugin
│   │   ├── github/               # GitHub plugin
│   │   └── my-command.md         # Custom command at root
│   ├── agents/                   # Agents (plugins + custom)
│   │   ├── core/                 # Core plugin
│   │   ├── github/               # GitHub plugin
│   │   └── my-agent.md           # Custom agent at root
│   ├── context/                  # Context (plugins + custom)
│   │   ├── core/                 # Core plugin
│   │   ├── lang-node/            # Lang-node plugin
│   │   └── my-context.mdc        # Custom context at root
│   └── avatars/                  # AI behavior profiles
│       └── my-avatar.md          # Custom avatar
│
├── .claude/                      # ❌ Generated (gitignored)
│   ├── CLAUDE.md                 # → .ai/AGENTS.md
│   ├── commands/                 # → .ai/commands/
│   ├── agents/                   # → .ai/agents/
│   ├── context/                  # → .ai/context/
│   └── output-styles/            # → .ai/avatars/
│
└── .cursor/                      # ❌ Generated (gitignored)
    ├── rules/                    # Directory with symlinks
    │   ├── main.mdc              # → .ai/AGENTS.md
    │   ├── core/                 # → .ai/context/core/
    │   ├── lang-node/            # → .ai/context/lang-node/
    │   └── *.md                  # → .ai/context/*.md
    ├── commands/                 # → .ai/commands/
    └── agents/                   # → .ai/agents/

# List all available plugins
.ai/cli plugins list

# Output:
#   ✓ core (installed)
#     github
#     code-quality
#     git
#     image-manipulation
#     lang-node
#     lang-typescript
#     lang-go
#     lang-ruby
#     lang-vue

# Install a plugin
.ai/cli plugins add lang-node

# Install multiple plugins
.ai/cli plugins add lang-typescript
.ai/cli plugins add github
.ai/cli plugins add code-quality

# Plugins are added to .ai/ and listed in .ai/config.jsonc

# Update agnostic-ai and all installed plugins
.ai/cli update

# Checks git status first (must be clean)
# Downloads latest version
# Re-installs your plugins
# Updates IDE configuration

curl -fsSL https://raw.githubusercontent.com/betagouv/agnostic-ai/main/install.sh | bash

Or if .ai folder is already installed, use .ai/cli configure

# Run from terminal
.ai/cli migrate

# This will show instructions for your IDE
# Then open your IDE (Claude Code or Cursor) and run:
/core:migrate

This command:

• Finds all documentation in your codebase
• Extracts relevant sections
• Organizes them into .ai/context/ files
• Removes extracted content from original files

# See what's available
.ai/cli plugins list

# Install language support
.ai/cli plugins add lang-node
.ai/cli plugins add lang-typescript

# Install GitHub integration
.ai/cli plugins add github

git add .ai/
git commit -m "feat: add AI configuration"
git push

Note: .ai/config.jsonc is committed and shared across the team!

# Configure IDE symlinks (can run multiple times to add more IDEs)
.ai/cli configure

# List available plugins
.ai/cli plugins list

# Install a plugin
.ai/cli plugins add <plugin-name>

# Examples
.ai/cli plugins add lang-node
.ai/cli plugins add github
.ai/cli plugins add code-quality

# Update agnostic-ai and installed plugins
.ai/cli update

# Install MCP server templates
.ai/cli mcp install

# List available MCP servers
.ai/cli mcp list

# Enable MCP server(s) for your IDE(s)
.ai/cli mcp use <server-name> [server-name...]

# Examples
.ai/cli mcp use github              # Enable GitHub integration
.ai/cli mcp use notion airtable     # Enable multiple servers
.ai/cli mcp use filesystem sqlite   # Local tools

# Disable MCP server(s) - Don't forget to disable unused servers to save tokens!
.ai/cli mcp unuse <server-name> [server-name...]

# Examples
.ai/cli mcp unuse github            # Disable GitHub when not working on PRs
.ai/cli mcp unuse notion airtable   # Disable multiple servers

💡 Tip: Only enable MCP servers you're actively using. Each enabled server consumes tokens in every conversation, so disable them when not needed to save costs!

Available MCP Servers (14 pre-configured):

• airtable - Airtable database operations
• context7 - Up-to-date library documentation
• filesystem - File system operations
• github - GitHub API integration
• gmail - Gmail search and drafts
• google-calendar - Calendar management
• google-drive - Drive, Sheets, and Docs access
• make - Make.com scenario integration
• next-devtools - Next.js development tools
• notion - Notion workspace integration
• postgres - PostgreSQL database access
• sqlite - SQLite database access
• stripe - Stripe payment processing
• trello - Trello board management

Setup: Add required API keys to .env.local (gitignored)

# Show help
.ai/cli help

Once installed, you have access to these commands:

Migrate existing documentation to .ai/context/ files (Run .ai/cli migrate from terminal first to see instructions)

Create a new slash command

Create a new specialized agent

Optimize and clean up context files

Perform deep research on a topic

Create a new AI personality/output style

Scaffold a new feature with EPCT methodology

# Discover available plugins
.ai/cli plugins list

# Install what you need
.ai/cli plugins add code-quality

# Commands appear instantly in your IDE
/code-quality:code-analyse

# Get latest updates
.ai/cli update

# Your installed plugins are automatically updated

# Pull changes
git pull

# Configuration is in .ai/config.jsonc - shared across the team
# If new plugins were added, they'll be automatically available

# Configure your IDE locally (each dev chooses their own IDE)
.ai/cli configure

Note: IDE configuration is local (gitignored) - each developer can use different IDEs!

Speed up your Claude CLI workflow with convenient aliases:

alias cc="claude --dangerously-skip-permissions"
alias ccc="claude --dangerously-skip-permissions -c"

Why it's safe: The agnostic-ai configuration includes a PreToolsBash hook that intercepts potentially destructive commands before execution, giving you speed without compromising safety.

📖 Full setup guide: templates/ides/claude/SHELL-SETUP.md

Want to add support for your favorite IDE or create a plugin?

See templates/ides/CONTRIBUTE.md for IDE integration guide.

For plugins, just create a folder in templates/plugins/ with your plugin name and structure:

templates/plugins/my-plugin/
├── commands/
│   └── my-command.md
├── agents/
│   └── my-agent.md
└── context/
    └── MY-CONTEXT.md

Original Idea & Development

This project was created by:

• Martin Ratinaud - Original concept and lead developer
• Maxime Dréau - Co-creator and architecture

Inspiration

This project was heavily inspired by @Melvynx's excellent aiblueprint. Thank you!

To get the most out of this project and AI-assisted development, we recommend Melvynx's AI Blueprint training to understand how to best work with AI assistants.

MIT

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Mattermost: Beta Gouv AI Channel

Made with ❤️ for developers who want to share AI configuration across all their AI coding tools