G-Rump a native macOS AI Coding Agent, looking for early feedback

Community Showcase
1

G-Rump, a native macOS AI coding agent — think Cursor or Windsurf, but built from scratch in pure Swift with zero external dependencies. I wanted to share it with the Swift community since it leans heavily on Apple frameworks and I'd love feedback.

What is it?

G-Rump is an AI-powered coding assistant that runs as a native macOS app. It has a full IDE-like interface with a sidebar, chat panel, and 17 specialized panels (file navigator, git, tests, terminal, profiling, localization, etc.). You chat with it, and it can read/write files, run shell commands, search the web, manage git, deploy to cloud, and more — 100+ tool definitions total.

Why Swift?

I wanted to prove that a full AI agent platform could be built natively on Apple's stack — no Electron, no Node.js runtime, no React Native. The result is an app that launches instantly, uses ~80MB of RAM idle, and integrates with macOS at a level that web-based tools simply can't. It has really pretty UI/UX as well.

| Category | Frameworks |

| **UI** | SwiftUI, AppKit, UIKit, CoreGraphics, QuartzCore, PDFKit, ImageIO |

| **AI/ML** | CoreML, NaturalLanguage, Vision, Speech |

| **System Integration** | CoreSpotlight, AppIntents, UserNotifications, ActivityKit, ScreenCaptureKit |

| **Security** | Security, CryptoKit, LocalAuthentication |

| **Data** | SwiftData, Foundation, Combine, UniformTypeIdentifiers |

| **Contacts & Calendar** | Contacts, EventKit |

| **Spatial** | ARKit, RealityKit (visionOS stubs) |

| **Media** | AVFoundation, CoreImage |

| **Diagnostics** | OSLog, os.signpost |

Key Technical Highlights

  • Deep Apple Integration

  • Spotlight — Conversations are indexed via `CoreSpotlight` so you can search your AI chat history from anywhere

  • Siri Shortcuts — `AppIntents` exposes "Ask G-Rump", "New Chat", and "Run Agent Task" to Shortcuts.app and Siri

  • Handoff — Start a conversation on your Mac, continue on another device

  • Live Activities — Long-running agent tasks show progress on the Lock Screen and Dynamic Island (iOS)

  • Secure Enclave — API keys stored via `LocalAuthentication` + Keychain, not plaintext

  • CoreML — On-device inference with downloadable quantized models from HuggingFace

  • NaturalLanguage — Semantic memory search using `NLEmbedding` for RAG (retrieval-augmented generation)

  • Focus Filters — Integrates with Focus modes to auto-configure agent behavior

Architecture

  • Zero external dependencies — `Package.swift` has an empty `dependencies: []` array. Everything is Apple frameworks + hand-rolled networking.
  • MCP Protocol — Supports 58 pre-configured Model Context Protocol servers for extensibility
  • Multi-provider AI — Anthropic, OpenAI, Ollama, OpenRouter, and on-device CoreML — all via a unified streaming interface
  • Skills system — 40+ bundled `SKILL.md` files (SwiftUI patterns, async/await, Kubernetes, code review) plus custom per-project skills
  • SOUL.md — Global and per-project AI personality configuration
  • Agent modes— Chat, Plan, Build, Debate, Spec, Parallel — each tailors the system prompt and tool selection
  • Unified `os.Logger`— All logging goes through Apple's unified logging system with 10 subsystem categories, zero `print()` statements
  • Strict concurrency — `StrictConcurrency=targeted` with zero warnings

IDE Features

  • 17 built-in panels: File Navigator, Git, Tests, Assets, Localization, Profiling, Logs, Terminal, App Store Tools, Apple Docs, and more

  • LSP integration via SourceKit-LSP for live diagnostics

  • Themes: Light, Dark, and fun themes matching Cursor, ChatGPT, Claude, Gemini, and Kiro

  • Zen Mode, Activity Bar, customizable layout, full keyboard shortcuts

  • 250Hz internal update loop for buttery-smooth streaming

Stats

  • ~25,000 lines of Swift across 80+ files
  • 100+ tool definitions with real executors
  • 40+ bundled skills
  • 10 test files with ~40 tests
  • Builds in ~10 seconds on M-series Macs
  • macOS 14+ / iOS 17+ / visionOS ready

What I Learned

  • SwiftUI is production-ready for complex apps — but you need `LazyVStack`, `drawingGroup()`, and careful state management for performance with large message lists.
  • `os.Logger` is dramatically better than `print()` — structured logging with categories, persistence, and Console.app filtering. Every Apple app should use it.
  • `AppIntents` is underrated — exposing actions to Siri/Shortcuts took ~50 lines of code and makes the app feel deeply integrated.
  • `NLEmbedding` is free on-device vector search — no need for external vector databases for small-to-medium RAG workloads.
  • SPM works for large apps — but you'll want `#if` flags to handle SwiftData macro expansion differences between `swift build` and Xcode

This project did not exist last week. I haven’t even had my Macbook Pro a Week. This is as far as i’ve gotten in one week. I am just one 24 year old from Mississippi with no coding background, i’m tired of being in poverty.

If you have the time I would greatly appreciate any Feedback and Critiques. I’ve only been coding since November so there will be mistakes.

Here is my Github Repo.

Screenshot (new account can only do one will post a link to the rest)

MainWindowLight
MainWindowLight2880×1800 221 KB

Imgur Link: G-Rump MACOS Screenshots - Album on Imgur

1 Like