extension/README.md

Dev - Meta AI Development Platform 🚀

The ultimate meta AI development platform. Manage and run ALL AI tools (Claude, Codex, Gemini, OpenHands, Aider) in one unified interface with authentication, API management, and parallel execution.

What You Get

• 200+ LLMs via Hanzo AI - Every model in one unified API
• 4000+ MCP Servers - Install ANY MCP server with one command
• Symbol Search - Find classes, functions, variables across all projects
• Universal MCP Proxy - Auto-install via npm/uvx and proxy all calls
• 45+ Legendary Modes - Code like Carmack, think like Norvig
• Unlimited Memory - Vector/graph/relational/symbol search
• Browser Automation - Built-in Playwright for web tasks
• Team Collaboration - Shared context and credits

🚀 Quick Start - Get Running in 2 Minutes

# Clone and setup
git clone https://github.com/hanzoai/dev.git
cd dev
make setup

# Login to Hanzo AI (opens browser)
make login
# OR use existing API key
export HANZO_API_KEY=hzo_... # from iam.hanzo.ai

# You're ready! Run any AI tool:
dev run claude "implement a REST API"
dev run aider "fix the failing tests" --auto-commit
dev run openhands "analyze this codebase" --worktree

# Multi-agent workflows
dev workflow code-review  # Review your changes
dev multi "optimize this function" --coder claude --reviewer gemini

📦 Installation Options

Option 1: CLI Tool (@hanzo/dev) - Recommended for Quick Start

# Install globally
npm install -g @hanzo/dev

# Login with your Hanzo account
dev login

# Initialize in your project
dev init

Option 2: VS Code Extension (Hanzo AI)

# Install from marketplace
code --install-extension hanzo-ai.hanzo-ai

# Or install .vsix locally
code --install-extension hanzo-ai-*.vsix

Option 3: MCP Server (@hanzo/mcp)

# For Claude Desktop
npx @hanzo/mcp@latest

🎯 Core Features

🔐 Unified Authentication & API Management

# Login once, use everywhere
dev login

# All your API keys are synced and encrypted locally
# - OpenAI/Codex API keys
# - Anthropic/Claude API keys  
# - Google/Gemini API keys
# - Auto-passthrough to all tools

🤖 Run Any AI Tool

# Claude - Advanced reasoning and coding
dev run claude "refactor this authentication system"

# Aider - Git-aware pair programming
dev run aider "add test coverage" --auto-commit

# OpenHands - Autonomous software engineering
dev run openhands "implement user management" --worktree

# Codex - Code generation
dev run codex "generate a REST API client"

# Gemini - Multimodal AI
dev run gemini "analyze this architecture diagram"

⚡ Async Long-Running Tasks

# Start task in background (auto-quits after 5min idle)
dev run claude "migrate database to PostgreSQL" --async
# Output: Job ID: abc123...

# Check status
dev status abc123

# Keep alive and send more instructions
dev input abc123 "also add connection pooling"

🌳 Parallel Development with Git Worktrees

# Spawn multiple AI agents working in parallel
dev run claude "implement auth" --worktree
dev run aider "add tests" --worktree  
dev run openhands "write docs" --worktree

# Each runs in its own branch and directory
dev worktree list

🔄 Compare AI Tools

# See how different AIs approach the same problem
dev compare "optimize this database query"

# Output shows results from all tools side-by-side

🤝 Multi-Agent Workflows

# Code Review with Multiple Perspectives
dev review  # Reviews current git changes
dev review src/api.js src/auth.js  # Review specific files

# Run Predefined Workflows
dev workflow code-review "review this PR #123"
dev workflow implement-feature "add user authentication"
dev workflow optimize "improve database query performance"
dev workflow debug "fix memory leak in production"

# Custom Multi-Agent Tasks
dev multi "design a REST API" --coder claude --reviewer gemini --critic codex

# Use Local LLMs (Ollama, LM Studio, etc)
dev multi "refactor this code" --local llama3 --reviewer gemini

🏠 Local LLM Support

# Auto-detects local LLMs (Ollama, LocalAI, etc)
dev run local-llm "explain this code" --model llama3

# Configure custom endpoints
dev config local-llm --add my-server http://localhost:8080

# Mix local and cloud models
dev multi "implement feature" --coder local:codellama --reviewer claude

🛠️ Local Development

# Quick setup with Make
make setup    # Install everything
make dev      # Start dev mode
make test     # Run tests

# Manual setup
npm install
cd packages/dev && npm install && npm link
cd packages/mcp && npm install

# Run locally
dev --help

Build from Source

# Build everything
make build

# Or individually:
npm run compile           # VS Code extension
cd packages/dev && npm run build  # CLI tool
cd packages/mcp && npm run build  # MCP server

CI/CD

All extensions are automatically built and tested on push:

• VS Code extension tests run on Node.js 18
• JetBrains plugin tests run on Java 17
• Releases are created automatically for tagged versions

🏗️ Architecture

• Dev CLI (@hanzo/dev) - Command-line interface for all AI tools
• Multi-Agent Orchestrator - Intelligent task routing and parallel execution
• Local LLM Manager - Seamless integration with Ollama, LocalAI, etc
• Hanzo AI Extension - VS Code/JetBrains integration
• MCP Server (@hanzo/mcp) - Model Context Protocol tools
• Platform Sync - Universal context and bi-directional file sync
• Async Wrapper - Long-running task management with idle detection

🧠 Intelligent Agent Assignment

Dev automatically assigns the right AI tool for each role:

• Claude - Architecture, complex reasoning, code review synthesis
• Gemini - Code review, documentation, multimodal tasks
• Codex - Code generation, optimization, critiques
• Aider - Git-aware coding, automated commits
• OpenHands - Autonomous feature implementation
• Local LLMs - Privacy-sensitive tasks, rapid iteration

🔗 Links

🚀 Login to Hanzo AI | 🌐 Hanzo AI | 📖 Docs | 💬 Discord

---

Built with ❤️ for engineers who ship fast.