Personal Assistant (macOS)

A local-first macOS menu bar assistant built with SwiftUI + AppKit.

It talks directly to LM Studio over its OpenAI-compatible API, supports streaming responses, and can call built-in and MCP tools.

Requirements

• macOS 26+
• Swift 6.2 toolchain
• LM Studio running with local server enabled at http://127.0.0.1:1234

Quick start

# from repo root
swift build
.build/debug/PersonalAssistant

Configuration

At startup, the app reads config from:

~/.config/pa/config.json

Seed it from the repository template:

mkdir -p ~/.config/pa
cp config.json.example ~/.config/pa/config.json

You can configure:

• LM Studio port
• generation defaults (max_tokens, temperature, system_prompt)
• MCP servers under mcp_servers

Usage

• Open from the menu bar icon.
• Global shortcut: Fn (Globe) + Control
• Press Esc to hide the assistant panel.

Architecture overview

• src/AppDelegate.swift
  App lifecycle, menu bar item, global hotkey registration, LM Studio status polling, MCP startup.
• src/AssistantWindow.swift + src/AssistantView.swift
  Floating panel + SwiftUI chat UI.
• src/BackendClient.swift
  Streaming chat loop, model discovery via /v1/models, tool-call handling via /v1/chat/completions.
• src/BuiltinTools.swift
  Built-in tools (get_current_datetime, get_weather).
• src/MCPManager.swift + src/MCPServer.swift
  MCP server process management + JSON-RPC stdio client.

Notes

• This repository is Swift-only (no Python backend layer).
• There is currently no first-party test/lint suite checked in.