Getting Started

AI-First is a CLI tool that prepares any repository to be used effectively by AI coding agents. It generates instant context so AI agents understand your codebase in seconds, not minutes.

Why AI-First?

AI coding assistants often struggle with large repositories due to:

• Limited context windows - Can't read all files
• Lack of architectural awareness - Don't understand project structure
• Difficulty navigating - Can't find relevant code
• Missing relationships - Don't understand module dependencies

AI-First solves this by creating a repository intelligence layer that provides:

• Project structure and architecture
• Extracted symbols (functions, classes, interfaces)
• Dependency graph
• Coding conventions
• Entry points

Quick Start

# Install globally (recommended)
npm install -g ai-first

# Or run without install
npx ai-first init

Generate Context

# Initialize AI-First in your repository
ai-first init

# This generates:
# - ai/ai_context.md      # Unified context (start here!)
# - ai/repo_map.json      # Machine-readable structure
# - ai/symbols.json       # Extracted functions/classes
# - ai/dependencies.json   # Import relationships
# - ai/architecture.md    # Architecture pattern
# - ai/tech_stack.md      # Languages & frameworks

Index for Fast Queries

# Create SQLite index for fast symbol queries
ai-first index

# For large repos, enable semantic search
ai-first index --semantic

Use with AI Agents

After running ai-first init, give the AI agent the ai/ai_context.md file. It contains:

• Project overview
• Architecture pattern
• Tech stack
• Entry points
• Key conventions

Supported AI Agents

Agent	How to Use
OpenCode	~/.config/opencode/commands/ai-first.md
Cursor	Reference ai/ai_context.md in prompts
Claude Code	Include context in system prompt
Windsurf	Project understanding
GitHub Copilot	Context-aware suggestions

Next Steps

• Installation Guide - More installation options
• Quick Start Tutorial - Step-by-step walkthrough
• Architecture - How AI-First works
• Examples - Real-world usage examples