PK Code CLI

PK Code is a powerful AI-driven command-line interface that transforms how developers interact with code. Built as a modern terminal application, it combines the power of large language models with intuitive developer workflows to streamline coding, debugging, and project management tasks.

✨ Key Features

• 🧠 Intelligent Code Analysis - Understand complex codebases instantly with AI-powered insights
• ⚡ Interactive Terminal Interface - Beautiful, responsive CLI built with React and Ink
• 🔗 Multi-Provider Support - Works with OpenAI, Anthropic, Google Gemini, and more
• 🎯 Context-Aware Assistance - Maintains project context across conversations
• 🛠️ Workflow Automation - Automate repetitive development tasks
• 🔍 Vision Model Integration - Advanced UI analysis and screenshot interpretation
• 📦 Monorepo Architecture - Scalable codebase with modular packages

🚀 Quick Start

Prerequisites

• Node.js 20+
• Your preferred AI provider API key

Installation

Verify installation

• Check the installed version on npm:
  • npm view pk-code-cli version
• Confirm the CLI shows the new flag and runs with a file prompt (safe no-op):
  • echo "Test" > prompt.txt (or use PowerShell Set-Content)
  • pk --prompt-file prompt.txt --list-extensions

From npm (Recommended)

npm install -g pk-code-cli
pk --version

From Source

git clone https://github.com/kingkillery/pk-code.git
cd pk-code
npm install
npm run build
npm install -g .

Configuration

Create a .env file in your project root or set environment variables:

OpenAI

export OPENAI_API_KEY="your_openai_key"
export OPENAI_MODEL="gpt-4"

Anthropic Claude

export ANTHROPIC_API_KEY="your_anthropic_key"
export ANTHROPIC_MODEL="claude-3-5-sonnet-20241022"

Google Gemini

export GOOGLE_API_KEY="your_google_key"
export GOOGLE_MODEL="gemini-1.5-pro"

🎯 Usage

Interactive Mode

Launch PK Code in interactive mode for conversational coding assistance:

cd your-project/
pk

Example interactions:

> Explain the architecture of this codebase
> Help me debug this React component
> Generate unit tests for the auth service
> Refactor this function to use TypeScript

Direct Commands

Execute specific tasks without entering interactive mode:

# Code generation
pk generate "Create a REST API endpoint for user authentication"

# Code analysis (use interactive prompt or provider-specific commands)
pk "What are the performance bottlenecks in this code?"

# Documentation (ask directly)
pk "Generate JSDoc comments for all functions in src/utils"

🔌 Supported AI Providers

| Provider | Models | Features | | -------------- | ------------------------------ | -------------------------------- | | OpenAI | GPT-4, GPT-3.5 | Chat, Code, Vision | | Anthropic | Claude 3.5 Sonnet, Claude 3 | Advanced reasoning, Code | | Google | Gemini 1.5 Pro, Gemini 1.0 Pro | Multimodal, Long context | | OpenRouter | Multiple models | Model variety, Cost optimization | | Cohere | Command R+, Command | Multilingual, RAG |

🎨 Vision & Multimodal Support

PK Code features advanced vision capabilities for UI analysis, screenshot interpretation, and visual debugging.

Features

• 🔄 Smart Model Routing - Automatically chooses between text and vision models
• 🖥️ UI Analysis - Specialized models for interface understanding
• 📸 Screenshot Processing - Analyze application screenshots and mockups
• 🔧 Browser Integration - Works seamlessly with browser automation tools

Configuration

# Enable vision capabilities
ENABLE_VISION_ROUTING=true
VISION_MODEL_NAME="gpt-4-vision-preview"
VISION_MODEL_PROVIDER="openai"
VISION_ROUTING_STRATEGY="auto"

Example Usage

# Analyze a screenshot
pk "What UI improvements can you suggest for this dashboard?"

# Debug visual issues
pk "Why is the layout broken on mobile devices?"

# Generate code from mockups
pk "Convert this design mockup into React components"

🌐 Browser Automation

PK Code supports browser automation through two methods: the cloud-based Browser Use API and a local browser-use MCP server, providing flexibility for different use cases.

Features

• Dual Mode Support: Choose between cloud API or local browser automation
• Cloud-based Browser Control: Execute browser automation tasks through the Browser Use API
• Local Browser Agent: Run browser automation locally via MCP integration
• UI Interaction: Click, type, and read content from web pages
• Structured Output: Get results in specified JSON formats
• Task Monitoring: Real-time streaming of task execution steps
• Task Control: Pause, resume, or stop running tasks

Configuration

Cloud API Mode (Default)

To use cloud-based browser automation, set your Browser Use API key:

export BROWSER_USE_API_KEY="your-api-key-here"

Local Browser Mode

For local browser automation without cloud dependencies:

# Prevent cloud API conflicts when using local browser
export PK_PREFER_LOCAL_BROWSER=1

# Start the local browser agent
pk
> /browser-use local

Usage Examples

Cloud API Usage

> pk
> Use browser_use to go to google.com and search for AI news

Local Browser Agent Usage

# Set environment variable to prefer local browser
export PK_PREFER_LOCAL_BROWSER=1

# Start PK Code
> pk
> /browser-use local
Local browser agent is ready!
> Navigate to https://github.com/kingkillery/pk-code and tell me how many open issues there are.

Environment Variables

| Variable | Description | Required | |----------|-------------|----------| | BROWSER_USE_API_KEY | API key for cloud Browser Use service | For cloud mode | | PK_PREFER_LOCAL_BROWSER | Set to 1 to disable cloud API and use local browser only | For local mode |

Troubleshooting

• Authentication errors: If you see 401 errors when using local browser, ensure PK_PREFER_LOCAL_BROWSER=1 is set
• Port conflicts: The local browser agent uses port 3001 by default
• Agent management: Use pk agent stop browser to stop the local browser agent

🔌 MCP Server Integration

PK Code supports Model Context Protocol (MCP) servers to extend functionality with custom tools and external integrations. MCP servers can provide additional capabilities like database access, API integrations, or specialized workflows.

Configuration

MCP servers are configured in the .pk/settings.json file. This file can be located either:

• Globally: ~/.pk/settings.json
• Per-project: .pk/settings.json in your project root

Add an mcpServers section to your settings file:

{
  "mcpServers": {
    "myServer": {
      "command": "node",
      "args": ["server.js"],
      "cwd": "./mcp-server",
      "trust": false
    }
  }
}

Usage

Once configured, MCP tools become available automatically. Use the /mcp command to view status:

> /mcp

See the MCP Server documentation for detailed configuration options and examples.

🔧 Advanced Configuration

Provider-Specific Settings

# OpenRouter (Access to multiple models)
export OPENROUTER_API_KEY="your_openrouter_key"
export OPENROUTER_MODEL="anthropic/claude-3.5-sonnet"

# Cohere (Optimized for enterprise)
export COHERE_API_KEY="your_cohere_key"
export COHERE_MODEL="command-r-plus"

Advanced Options

# Context settings
export PK_MAX_CONTEXT_SIZE=32000
export PK_CONVERSATION_MEMORY=true

# Performance tuning
export PK_RESPONSE_TIMEOUT=30000
export PK_CONCURRENT_REQUESTS=3

# Debug mode
export DEBUG=pk:*

🚀 CI/CD Integration

GitHub Actions

Integrate PK Code into your CI/CD pipeline to run scripted prompts:

name: PK Code Prompt

on: [pull_request]

jobs:
  prompt:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Setup Node.js
        uses: actions/setup-node@v4
        with:
          node-version: '20'

      - name: Install PK Code
        run: npm install -g pk-code-cli

      - name: Run prompt
        env:
          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
        run: |
          pk "Review the changes in this PR for potential issues"

📚 Common Use Cases

🔍 Code Review & Analysis

# Analyze code quality
pk "Review this function for potential bugs and improvements"

# Security audit
pk "Check this code for security vulnerabilities"

# Performance analysis
pk "Identify performance bottlenecks in this module"

🏗️ Code Generation

# Generate boilerplate
pk "Create a complete CRUD API for a user management system"

# Generate tests
pk "Write comprehensive unit tests for the authentication service"

# Generate documentation
pk "Create detailed API documentation for all endpoints"

🐛 Debugging & Problem Solving

# Debug issues
pk "Help me understand why this React component isn't re-rendering"

# Explain error messages
pk "Explain this TypeScript error and suggest fixes"

# Architecture guidance
pk "Suggest the best design pattern for this use case"

📊 Performance & Benchmarks

| Task Type | Response Time | Accuracy | Token Efficiency | | --------------- | ------------- | -------- | ---------------- | | Code Analysis | 2-5s | 95% | ⭐⭐⭐⭐⭐ | | Code Generation | 3-8s | 92% | ⭐⭐⭐⭐ | | Debugging Help | 1-3s | 96% | ⭐⭐⭐⭐⭐ | | Documentation | 2-6s | 94% | ⭐⭐⭐⭐ |

🏗️ Architecture

PK Code is built with a modern, extensible architecture:

pk-code/
├── packages/
│   ├── core/              # Core engine and utilities
│   ├── cli/               # Command-line interface
│   ├── vscode-ide-companion/  # VS Code integration
│   └── shared/            # Shared types and utilities
├── docs/                  # Documentation and guides
├── examples/              # Usage examples
├── integration-tests/     # End-to-end tests
└── scripts/              # Build and deployment scripts

Key Components

• 🎛️ Core Engine - Handles AI provider communication and context management
• 💻 CLI Interface - React-based terminal UI built with Ink
• 🔌 Provider Adapters - Standardized interfaces for different AI services
• 🧠 Context Manager - Maintains conversation state and project context
• ⚡ Response Processor - Formats and validates AI responses

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Setup

# Clone the repository
git clone https://github.com/kingkillery/pk-code.git
cd pk-code

# Install dependencies
npm install

# Run tests
npm test

# Start development
npm run start

Running Tests

# Unit tests
npm run test

# Integration tests
npm run test:integration

# E2E tests
npm run test:e2e

🆘 Support & Troubleshooting

Common Issues

API Key Issues

# Verify your API key is set
echo $OPENAI_API_KEY

# Test connection
pk config test

Performance Issues

# Enable debug mode
export DEBUG=pk:*
pk your-command

Memory Issues

# Reduce context size
export PK_MAX_CONTEXT_SIZE=16000

Getting Help

• 📖 Documentation
• 🐛 Report Issues
• 💬 Discussions
• 📧 Email Support

📄 License

PK Code is licensed under the Apache License 2.0.

🌟 Acknowledgments

Built on the foundation of Google Gemini CLI. Special thanks to the Gemini CLI team for their excellent work that made this project possible.