Penguin Office

Introducing Penguin Office 🐧 An isometric 3D desktop app where your AI coding agents come to life as penguins in a virtual office. Penguin Office connects to your local AI CLI tools — Claude, Codex, Gemini, Droid, OpenCode, Ollama — and turns them into living agents inside an interactive workspace. Assign tasks via chat or kanban, then watch your penguins: • walk to their desks • write code • collaborate in real time • take breaks, exercise, and move around Each penguin represents a real running CLI process. Each one has its own role, outfit, and behavior. This isn’t a toy — it’s a working agent orchestration layer with a visual interface. CA : 0xCC280074e4Bd5167D5c2fB3e43524B9D2A0b6ca6 penguoffice.tech github.com/penguagent/pen…

Install and Run : git clone github.com/penguagent/pen… cd penguin-office pnpm install pnpm dev

How Our Agent Orchestration Works? Each CLI tool has an adapter that implements a common interface: interface AgentAdapter { name: string; detect(): Promise; spawn(task: Task, workDir: string): AgentProcess; kill(): void;} When you send a task: 1. The AgentManager selects an available adapter 2. It spawns the CLI as a child process with the task as an argument 3. stdout/stderr are streamed through the EventBus 4. Events are consumed by the SceneManager to drive penguin animations 5. Events are also consumed by the Zustand store to update UI panels The Electron main process handles actual process spawning via IPC. When running in browser dev mode (without Electron), a simulation mode generates fake agent output so you can see the office working.

Click the Tasks button in the bottom-left corner to open the kanban board. Tasks flow through four columns: > Backlog -- queued tasks waiting for assignment > In Progress -- actively being worked on by an agent > Review -- completed but awaiting your review > Done -- finished tasks Each card shows the assigned penguin's avatar, the task description, and token usage.

Use the Chat Panel on the right side of the screen. Type a message to send it to the next available agent, or use @mentions to target a specific one: claude refactor the authentication module to use JWT tokens, codex write unit tests for the payment service, gemini analyze this codebase and suggest performance improvements ollama explain how the routing layer works When a task is sent: - The penguin walks to its assigned desk - It sits down and starts typing (flippers animate) - A speech bubble shows progress from the agent's stdout The status dot turns green (working) When the task completes, the penguin stands up and walks to the break room If an error occurs, the status dot turns red and the bubble shows the error penguoffice.tech

The app auto-detects which CLIs you have installed. Only installed agents appear as penguins in the office. You can use any combination.

$POT Dex is Paid! what if your dev team was just… agents? - you assign tasks - they walk to their desks - write code - open PRs - ship products Penguin Office + @gitlawb starts to look like that. dexscreener.com/base/0x72418ba…

Click the Logs button to see raw agent output streaming in real time. Useful for debugging or monitoring what each agent is actually producing. penguoffice.tech

Penguin Office is An isometric 3D desktop app where your AI coding agents come to life as penguins in a virtual office. Penguin Office is an agentic orchestration application that connects to your locally installed CLI AI tools -- Claude, Codex, Gemini, Droid, OpenCode, and Ollama -- and visualizes them as penguin characters working in a fully interactive office environment. Assign tasks through chat or a kanban board, and watch your penguins walk to their desks, type at computers, take breaks, exercise, and collaborate in real time. Each penguin wears a unique outfit based on the AI it represents. Each one has a real CLI process running behind it. This is not a toy -- it is a working orchestration layer with a visual interface.

Connect your local CLI tools — Claude, Codex, Gemini, Ollama, and more — and watch them turn into penguins working inside a virtual office. Assign tasks → they walk, sit, type, and collaborate in real time. Each penguin = a real running agent. Chat, kanban, logs — all connected to actual execution. This is what agent orchestration looks like when it becomes visible.

Penguin Office is build for Distribution.

The app auto-detects which CLIs you have installed. Only installed agents appear as penguins in the office. You can use any combination.

Each penguin wears a unique outfit based on the AI it represents. Each one has a real CLI process running behind it. This is not a toy -- it is a working orchestration layer with a visual interface.

A compact, always-on-top transparent overlay that sits on your desktop. Shows only the office scene without panels. Toggle between modes with: Keyboard shortcut: Cmd+Shift+P(Mac) / Ctrl+Shift+P(Windows/Linux) System tray: Right-click the tray icon and select "Toggle Widget Mode" In widget mode, the penguins live on your desktop while you work.

soon you’re able to talk with our agents on @X!

The app auto-detects which CLIs you have installed. Only installed agents appear as penguins in the office. You can use any combination. explore more on : penguoffice.tech

Penguin Office furniture: desks with monitors, couches, ping pong table, treadmills, server racks, water cooler, plants Floating name labels and colored status dots (green = working, yellow = idle, red = error) Speech bubbles showing real-time agent output Dark UI panels overlaid on the scene.

When a task is sent: 1The penguin walks to its assigned desk 2It sits down and starts typing (flippers animate) 3A speech bubble shows progress from the agent's stdout 4The status dot turns green (working) 5When the task completes, the penguin stands up and walks to the break room 6If an error occurs, the status dot turns red and the bubble shows the error

Use the Chat Panel on the right side of the screen. Type a message to send it to the next available agent, or use mentions to target a specific one: - claude refactor the authentication module to use JWT tokens - codex write unit tests for the payment service - gemini analyze this codebase and suggest performance improvements - ollama explain how the routing layer works