dot-ai

Generate AI provider configurations from a centralized .ai/ folder

A CLI tool that reads from a single .ai/ directory and generates configuration files for multiple AI assistants and code editors, eliminating duplicate configs across Claude Code, Cursor, Gemini, and OpenCode.

The Problem

Modern development often involves multiple AI assistants and code editors, each requiring their own configuration files:

• Claude Code needs CLAUDE.md
• Cursor needs .cursor/rules/*.mdc files
• Gemini needs GEMINI.md and .gemini/settings.json
• Codex needs AGENTS.md
• OpenCode needs opencode.json
• MCP servers need .mcp.json

This leads to scattered configs and duplicate content across your project.

The Solution

dot-ai provides a single source of truth in a .ai/ folder that generates all provider-specific configurations:

.ai/ (single source)
├── instructions.md       # Main instructions
├── rules/               # Rules with YAML frontmatter
│   ├── general.md
│   ├── typescript.md
│   └── forms.md
├── commands/            # Available commands
│   ├── deploy.md
│   └── test.md
└── mcp.json            # MCP server configuration

↓ Generates ↓

CLAUDE.md                 # Claude Code instructions
GEMINI.md                 # Gemini instructions (identical to Claude)
AGENTS.md                 # Agents instructions (identical to Claude)
.mcp.json                 # MCP server config
.cursor/rules/*.mdc       # Cursor rules (preserves frontmatter)
.gemini/settings.json     # Gemini MCP settings
opencode.json            # OpenCode MCP config

Before and After

Important Note: This is a Temporary Solution

dot-ai is intended as a stop-gap solution. The ultimate goal is for AI model providers and development tools to standardize on a single configuration format, so every provider and CLI tool reads from the same place. When that happens, this project will be deprecated.

Until that standardization happens, dot-ai helps eliminate the pain of maintaining duplicate configurations across multiple tools. We hope this project becomes obsolete as the ecosystem matures and converges on unified standards.

Running the CLI

# If you have bun installed
bunx dot-ai@latest

# If you don't have bun installed
npx bun x dot-ai@latest

Usage

Quick Start

# Initialize .ai/ folder structure or migrate existing configs (one-time setup)
bun dot-ai@latest init

# Generate all AI provider configs from .ai/ folder
bun dot-ai@latest run

Commands

• init - Initialize .ai/ folder structure or migrate existing configs
• run - Generate provider-specific configs from .ai/ folder

The run command will:

1. Read your .ai/ folder structure
2. Generate provider-specific configuration files
3. Preserve YAML frontmatter where needed (Cursor)
4. Transform MCP configs to each provider's format

.ai/ Folder Structure

instructions.md

Main instructions that will be included in CLAUDE.md, GEMINI.md, and AGENTS.md:

# Project Instructions

Follow these guidelines when working on this project...

rules/ Directory

Rule files with YAML frontmatter that define specific coding standards:

---
title: TypeScript Rules
enabled: true
priority: 1
---

# TypeScript Guidelines

- Always use explicit types
- Avoid `any` type
- Use strict mode

commands/ Directory

Documentation for available commands (included in instruction files):

# Deploy Command

Handles deployment to production environments.

## Usage

mcp.json

MCP server configuration that gets distributed to all providers:

{
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "npx",
      "args": ["@modelcontextprotocol/server-filesystem", "/tmp"]
    }
  }
}

Generated Files

File	Description	Content
CLAUDE.md	Claude Code instructions	instructions + rules (no frontmatter) + commands
GEMINI.md	Gemini instructions	Identical to CLAUDE.md
AGENTS.md	Agents instructions	Identical to CLAUDE.md
.mcp.json	MCP server config	Direct copy of .ai/mcp.json
.cursor/rules/*.mdc	Cursor rules	Individual rule files with frontmatter preserved
.gemini/settings.json	Gemini MCP settings	{ "mcpServers": { ... } }
opencode.json	OpenCode MCP config	{ "mcp": { "server": { "type": "local", "command": [...] } } }

How to Migrate to .ai

If you have existing AI configs, follow this workflow to safely migrate:

1. Create a Migration Branch

git checkout -b migrate-to-dot-ai

2. Run Migration

bun dot-ai@latest init

This creates a .ai/ folder by consolidating your existing configs:

• CLAUDE.md, GEMINI.md, AGENTS.md → .ai/instructions.md
• .cursor/rules/*.mdc → .ai/rules/*.md
• Various MCP configs → .ai/mcp.json

3. Review and Clean Up .ai/ Folder

Important: The migration may concatenate multiple files or create duplicate MCP configs. Edit the files as needed to remove duplicates and organize content properly.

4. Delete Original AI Config Files

Once satisfied with .ai/ folder, remove the original AI-specific files:

• Always remove: CLAUDE.md, GEMINI.md, AGENTS.md, .cursor/rules/
• MCP configs: Remove .mcp.json, .gemini/settings.json, opencode.json since these will be auto-generated
• Keep other configs: Don't remove .json files that serve purposes beyond MCP server configuration (like editor settings, build configs, etc.)

5. Commit the Migration

Commit the current work so we have a checkpoint.

6. Generate New Configs

bun dot-ai@latest run

7. Add Generated Files to .gitignore

You can add the generated files to .gitignore if you prefer.

8. Re-generate when needed

You can now use bun dot-ai@latest run whenever you update .ai/ configs.

Starting Fresh

If you don't have existing AI configs, initialize a new .ai/ folder by running:

bun dot-ai@latest init

This will create the .ai/ folder structure and example files to get you started.

Contributing

Contributions are welcome! This project aims to:

1. Support more providers as they emerge
2. Improve configuration mapping between formats
3. Eventually become unnecessary once standards emerge

Running the Project Locally

# Install dependencies
bun install

# Build the project
bun run build

# Test the built CLI
bun link
bun link dot-ai

# Test in another project by running the linked cli
cd ~/my-project
bunx dot-ai run

Running Tests

# Run all tests
bun test

# Run type checking
bun typecheck

Please open issues for bugs or feature requests.

License

MIT License - see LICENSE for details.