MLX Code v6.1.0 — Security Hardening & Code Quality Audit

What's New in v6.1.0

Security Hardening & Code Quality Audit

31 findings resolved across CRITICAL, HIGH, MEDIUM, LOW, and INFO severities.

Critical Fixes:

• API Keys to Keychain: All AI backend API keys migrated from UserDefaults to macOS Keychain with automatic migration on first launch

High Fixes:

• Command Validator Hardened: NSRegularExpression word-boundary matching prevents bypass via substrings
• Python Import Validator: Regex-based validation with comment filtering
• Model Hash Verification: SHA256 verification using CryptoKit
• Buffered I/O: 4096-byte chunk reading replaces byte-by-byte daemon communication
• Task Cancellation: while !Task.isCancelled for clean shutdown
• Portable Paths: Bundle-relative paths replace hardcoded file paths
• Secure Logging: SecureLogger replaces all print() statements

Medium Fixes:

• Unicode search with localizedCaseInsensitiveContains()
• O(n) context management replacing O(n²) insert-at-zero
• 1MB file content cap, multi-version Python path lookup
• Serial queues for thread-safe MLX service operations
• Permission checks for script execution

Low/Info Fixes:

• Force unwrap elimination, NSString cast → URL API
• Named constants for context budget ratios
• Clear Conversations confirmation dialog
• Deprecated unused ContentView with @available attribute

Install

Download the DMG below, open it, and drag MLX Code to Applications.

Requirements: macOS 14.0+, Apple Silicon (M1/M2/M3/M4), 8GB RAM minimum

MLX Code v1.3.0

MLX Code v1.3.0

Local LLM-powered coding assistant for macOS using Apple's MLX framework. Privacy-first alternative to GitHub Copilot — no cloud, no telemetry, runs entirely on Apple Silicon.

Features

• Code completion and generation using local LLMs
• Chat-based coding assistant
• Multi-model support (MLX, Ollama, OpenAI-compatible)
• Code explanation and refactoring suggestions
• Privacy-first: all inference runs locally on Apple Silicon
• No internet connection required for local models

Installation

1. Download MLX-Code-v1.3.0-Final.dmg below
2. Open the DMG and drag MLX Code to Applications
3. Launch MLX Code

Requirements

• macOS 14.0 (Sonoma) or later
• Apple Silicon Mac (M1/M2/M3/M4)
• MLX (pip install mlx-lm) or Ollama (brew install ollama)

What's New in v1.3.0

• Enhanced cloud AI integration
• Improved model management
• Performance optimizations
• Bug fixes

Created by Jordan Koch

MLX Code v1.0.0 - Initial Release

MLX Code v1.0.0 - Initial Release

🚀 Local AI-Powered Coding Assistant for macOS

A privacy-first alternative to cloud-based AI coding tools, running entirely on your Mac using Apple's MLX framework.

✨ Key Features

Core Capabilities

• Local AI Execution: Complete privacy - no data leaves your machine
• MLX Framework: Apple Silicon optimized using MLX for fast inference
• Xcode Integration: Build, test, and analyze projects directly
• Chat Interface: Claude Code-style conversational interface
• File Operations: Read, write, edit, and search files with AI assistance

Advanced Features

• 20 Built-in Templates: Code generation, refactoring, documentation, debugging
• Git Integration: Smart commit messages and repository operations
• Keyboard Shortcuts: 15+ shortcuts for productivity
• Markdown Rendering: Syntax-highlighted code blocks
• Build Error Parser: AI-powered fix suggestions
• Multi-Conversation: Manage multiple coding sessions
• Secure Storage: Keychain integration for credentials

Supported Models

• Deepseek Coder 6.7B (Recommended) - Fast and accurate
• CodeLlama 13B - Strong reasoning
• Qwen Coder 7B - Balanced performance
• Custom Models - Add any MLX-compatible model

📊 Project Stats

• Lines of Code: 8,500+
• Swift Files: 29
• Features: 50+
• Templates: 20 built-in
• Keyboard Shortcuts: 15+

📥 Installation

Prerequisites

1. macOS 14.0+ (Sonoma or later)
2. Apple Silicon (M1/M2/M3/M4 recommended)
3. Python 3.10+

Python Setup

# Create virtual environment
python3 -m venv ~/mlx-env
source ~/mlx-env/bin/activate

# Install MLX
pip install mlx mlx-lm numpy transformers

Install MLX Code

1. Download MLXCode-v1.0.0-macOS.tar.gz
2. Extract the archive
3. Copy MLX Code.app to /Applications
4. Launch and configure Python path in Settings

🎯 Quick Start

1. Set Python Path: Settings → Advanced → Python interpreter path
2. Select Model: Choose from pre-configured models or add custom
3. Start Chatting: Ask coding questions, request file operations, build projects

Example Commands

"Read the contents of ViewController.swift"
"Create a new NetworkManager class with URLSession"
"Build the current project and fix any warnings"
"Write unit tests for this function"
"Generate a commit message for staged changes"

⌨️ Keyboard Shortcuts

• ⌘N - New conversation
• ⌘K - Clear conversation
• ⌘⇧T - Template library
• ⌘⇧G - Git helper
• ⌘⇧B - Build project
• ⌘/ - Command palette
• ⌘, - Settings

🔒 Security & Privacy

• ✅ 100% Local Processing - No cloud dependencies
• ✅ Sandboxed Execution - macOS app sandbox
• ✅ Input Validation - Protection against injection attacks
• ✅ Secure Storage - Keychain for sensitive data
• ✅ No Telemetry - Zero data collection
• ✅ Path Traversal Protection - Safe file operations

🏗️ Architecture

• Language: Swift 5.9+
• UI: SwiftUI
• ML: MLX (via Python bridge)
• Pattern: MVVM with Combine
• Concurrency: Actor-based
• Security: Input validation, sandboxing, encryption

📚 Documentation

• README.md - Complete guide
• FEATURES.md - Feature details
• SECURITY.md - Security documentation
• PROJECT_SUMMARY.md - Technical overview

⚡ Performance

Benchmarks (M2 Max, 32GB)

• Model Load: ~10s (7B 4-bit)
• First Token: ~2s
• Tokens/sec: ~45
• Memory: 6-10GB (depending on model)

🔧 Requirements

Minimum

• M1 Mac (8GB RAM)
• macOS 14.0
• 10GB free disk
• 7B quantized model

Recommended

• M2 Pro/Max or M3 (16GB+ RAM)
• macOS 14.0+
• 50GB free disk
• Multiple models cached

🐛 Known Limitations

• Requires manual Python/MLX setup
• Models must be downloaded separately
• macOS only (by design)
• Requires Command Line Tools

🔮 Future Enhancements

• Context-aware project analysis
• Interactive diff viewer
• Advanced debugging integration
• Custom model fine-tuning
• Plugin system

📜 License

MIT License - See LICENSE file

👥 Authors

Jordan Koch & Claude Code

---

Platform: macOS 14.0+
Architecture: Apple Silicon optimized
Version: 1.0.0
Status: Production-ready