Worth Doing

AI-Powered Desktop Agent for macOS
Describe a task in natural language. The agent plans it, you approve it, it executes it.

---

What is Worth Doing?

Worth Doing is a native macOS application that turns natural language instructions into real desktop automation. It uses OpenRouter LLMs for reasoning and planning, then executes actions through shell commands and AppleScript — step by step, with full visibility and safety controls.

How it works

You type: "Create a project folder on my Desktop with subfolders for docs, images, and exports"

The agent:
  1. Interprets your request
  2. Builds a structured execution plan (shell commands, AppleScript, file ops)
  3. Shows you the plan for review
  4. You approve → it executes each step sequentially
  5. You see live output, errors, retries, and final result

---

Key Features

Agent Execution Engine Sequential step-by-step execution 10 step types: shell, AppleScript, file I/O, LLM reasoning, verification, and more Automatic error recovery via LLM step repair Pause / Resume / Cancel mid-execution Live execution timeline with output inspection	Safety & Transparency Risk classification engine (low / medium / high) Safe Mode: medium + high risk actions require approval Command-level pattern analysis (detects rm -rf, sudo, etc.) Full visibility into every command before and after execution macOS permission status panel
LLM Planning OpenRouter integration (any model) Structured JSON plan generation Task interpretation → assumptions → risk assessment → steps Step repair on failure (asks LLM to fix broken commands) Model picker with context length and pricing info	Workflows & History Save successful executions as reusable workflows Favorite workflows for quick access Full task history with search and filtering Session logging with step-by-step results Re-run workflows with one click

---

Architecture

Sources/
├── BrainOS/           App entry point, state, navigation
├── Core/              Models, persistence (JSON), safety engine
├── DesignSystem/      Theme, glass cards, visual effects, badges
├── SecureStorage/     Keychain wrapper (API key storage)
├── AIService/         OpenRouter client (models, completions)
├── Agent/             Coordinator, PlanBuilder, StepRunner, UI
├── Executors/         ShellExecutor, AppleScriptExecutor
├── Dashboard/         Main view, task composer
├── Settings/          API key, model picker, permissions, safety
├── History/           Task history with search/filter
├── Workflows/         Saved workflow library
└── Onboarding/        First-launch setup flow

Agent State Machine

idle → planning → awaitingApproval → executingStep → completed
                        ↓                  ↓
                   (user rejects)     verifyingStep
                    → cancelled       recoveringStep
                                         paused
                                        → failed

---

Tech Stack

Component	Technology
Language	Swift 5.9+
UI Framework	SwiftUI (macOS 14+)
Build System	Swift Package Manager
LLM Backend	OpenRouter API
Shell Execution	Process (stdin/stdout/stderr pipes)
AppleScript	NSAppleScript
Secrets	macOS Keychain (Security framework)
Persistence	JSON files (~/Library/Application Support/BrainOS/)
Signing	Developer ID + Apple Notarization
Dependencies	Zero external dependencies

---

Metrics

---

Install

Download (Recommended)

1. Download the latest WorthDoing.dmg from Releases
2. Open the DMG and drag Worth Doing to Applications
3. Launch — the app is signed and notarized by Apple, no Gatekeeper warning

Build from Source

git clone https://github.com/Worth-Doing/brain-os.git
cd worth-doing
swift build -c release
swift run BrainOS

Requires macOS 14+ and Swift 5.9+ (Xcode Command Line Tools). No Xcode project needed.

Create .app Bundle

swift build -c release
bash Scripts/bundle-app.sh
open ".build/release/Worth Doing.app"

---

Setup

1. Get an OpenRouter API key at openrouter.ai
2. Launch the app — the onboarding will guide you through:
   • Entering your API key (stored in macOS Keychain)
   • Selecting a default model
3. Start describing tasks on the Dashboard

---

Example Tasks

Task	What the agent does
"Create a folder on Desktop called Research with subfolders PDFs, Notes, Exports"	mkdir -p commands, verifies structure
"List all PNG files in Downloads and rename them receipt-001, receipt-002..."	find + sequential mv commands with preview
"Open Safari"	AppleScript tell application "Safari" to activate
"Find files larger than 100MB in my home directory"	find ~ -size +100M with formatted output
"Create a markdown report of my system info"	Runs sw_vers, sysctl, df and writes .md file

---

Safety Model

Risk Level	Examples	Behavior
	Read files, list dirs, open apps	Executes automatically
	Move/rename files, write files, app automation	Requires approval in Safe Mode
	Delete files, sudo, rm -rf, bulk operations	Always requires approval

The RiskEngine scans every command against 35+ known dangerous patterns before execution.

---

License

MIT License. See LICENSE for details.

---

_{Built with Swift + SwiftUI. No Xcode. No Electron. No web wrappers. Just native macOS.}