A browser-based AI coding IDE where every task runs through an autonomous loop.
Wiggum is a fully browser-based AI coding assistant that works autonomously. Unlike traditional AI chat interfaces where you go back and forth, Wiggum takes your task and runs with it — iterating automatically until the job is done.
Oops! It's All Loops.
Every message you send goes through the Ralph loop:
- Simple tasks (questions, small fixes) complete in one iteration
- Complex tasks (build an app, refactor a codebase) run multiple iterations with fresh context each time
- The AI signals when it's done — you don't have to babysit it
You: "Create a todo app with local storage"
↓
┌─────────────────────────────────────┐
│ Ralph Loop (Iteration 1) │
│ - Read task from .ralph/task.md │
│ - Plan approach │
│ - Create initial files │
│ - Update .ralph/progress.md │
└─────────────────────────────────────┘
↓
┌─────────────────────────────────────┐
│ Ralph Loop (Iteration 2) │
│ - Read progress from last round │
│ - Add styling and interactions │
│ - Test the code works │
│ - Update progress │
└─────────────────────────────────────┘
↓
┌─────────────────────────────────────┐
│ Ralph Loop (Iteration 3) │
│ - Read progress │
│ - Final polish │
│ - Write "complete" to status.txt │
└─────────────────────────────────────┘
↓
Done! Your todo app is ready.
Each iteration gets fresh context — the AI reads its task and progress from files, not from an ever-growing conversation. This means complex tasks don't hit token limits.
- Autonomous by Default — Every task runs through the Ralph loop automatically
- Fully Browser-Based — No server required; runs entirely in your browser
- In-Browser Filesystem — LightningFS with IndexedDB persistence
- Git Integration — Full git support via isomorphic-git
- Live Preview — In-browser code compilation with esbuild-wasm
- Tool-Using AI — File read/write, shell commands, and more
- Skills System — Extensible AI capabilities via skill definitions
- Node.js 18+
- pnpm 9+
# Clone the repository
git clone https://github.com/PerceptLabs/wiggum.git
cd wiggum
# Install dependencies
pnpm install
# Start the dev server
pnpm devOpen http://localhost:5173 in your browser.
- Open Wiggum in your browser
- Go to Settings and enter your OpenAI-compatible API key
- Start chatting — your tasks will automatically run through the loop
Ralph is the heart of Wiggum. It's the autonomous iteration system that handles every task.
- Task Capture — Your message becomes
.ralph/task.md - Iteration — AI reads task, makes progress, updates
.ralph/progress.md - Fresh Context — Each iteration starts clean, reading state from files
- Completion Signal — AI writes "complete" to
.ralph/status.txtwhen done
.ralph/
├── task.md # The original task
├── progress.md # What's been accomplished
├── status.txt # Current status (running/complete/waiting)
└── iteration.txt # Current iteration number
Traditional AI chat has a problem: as conversations grow, you hit context limits. The AI forgets earlier messages or gets confused.
Ralph solves this with a simple idea: don't accumulate context, iterate with fresh context. Each iteration:
- Reads the task (what to do)
- Reads progress (what's done)
- Does the next step
- Updates progress
This means a 50-iteration task works just as well as a 2-iteration task.
wiggum/
├── apps/
│ └── ide/ # Wiggum IDE application
│ ├── src/
│ │ ├── components/ # React components
│ │ ├── contexts/ # React context providers
│ │ ├── hooks/ # Custom React hooks
│ │ ├── lib/ # Core libraries
│ │ │ ├── ai/ # AI client and streaming
│ │ │ ├── build/ # esbuild-wasm integration
│ │ │ ├── commands/ # Shell command implementations
│ │ │ ├── fs/ # Filesystem abstraction
│ │ │ ├── session/ # Session management
│ │ │ ├── skills/ # Skills system
│ │ │ └── tools/ # AI tool definitions
│ │ └── pages/ # Route pages
│ └── package.json
├── packages/
│ └── stack/ # @wiggum/stack UI library
│ ├── src/
│ │ ├── components/ui/ # 53 shadcn/ui components
│ │ ├── hooks/ # Shared React hooks
│ │ └── lib/ # Utilities (cn, formatters)
│ └── package.json
├── pnpm-workspace.yaml
└── package.json
| Category | Technology |
|---|---|
| Framework | React 19, TypeScript |
| Build | Vite 7 |
| Styling | Tailwind CSS 4 |
| State | @tanstack/react-query |
| Filesystem | LightningFS, isomorphic-git |
| Bundler | esbuild-wasm |
| AI | OpenAI-compatible API |
| Category | Technology |
|---|---|
| Components | 53 shadcn/ui components |
| Primitives | Radix UI |
| Tables | @tanstack/react-table |
| Forms | react-hook-form, zod |
| Utilities | clsx, tailwind-merge |
Full documentation available in the docs/ folder:
MIT