Pixtuoid: Terminal Pixel-Art Office for AI Coding Agents

What changed

The release of Pixtuoid introduces a novel approach to monitoring and interacting with AI coding agents. Developed in Rust, this application transforms agent sessions into animated pixel-art characters occupying desks within a virtual terminal office. Each agent is represented by a character that visually indicates its status: typing when active, raising a question mark when awaiting input, and dozing off when completed. The office layout includes distinct zones like cubicles, a meeting lounge, and a pantry, with new floors automatically added as agent capacity is reached.

Key features include:

Visualizations and Status Indicators

• Animated Characters: Coworkers animate their actions, including typing, waiting, and sleeping, with paths routed between desks.
• Team Palette: Agent sessions share shirt and pant colors derived from the working directory, aiding in organizational clarity. Hair and skin tones vary per agent, with 16 curated outfit options.
• Per-Tool Monitor Glow: Desk monitors display colors corresponding to the tool in use (e.g., blue for Edit, orange for Bash, cyan for Read), allowing for quick identification of ongoing tasks.
• Token Meter: A visual representation of token usage is shown as paper stacks on desks, with tiers for 250K, 2M, and 16M tokens. Hovering reveals exact token counts.
• Hover Tooltips: Detailed information on session duration, tool call counts, and active time percentages is available upon hovering over an agent.
• Agent Tree Dashboard: Accessible via the Tab key, this dashboard presents a collapsible tree of all agents, color-coded by their current activity and displaying tool call counts.
• Office Pets and Mascots: A roaming cat or dog appears on each floor, and an OpenClaw gateway is visualized as a moving lobster, indicating gateway health.
• Dynamic Office Environment: The office features a day/night cycle, weather effects (rain, snow, fog), and six customizable themes.
• Lofi Soundtrack: A synthesized lofi soundtrack adjusts to the office's clock and weather, with typing sounds and environmental cues.

Installation and Integration

Pixtuoid can be installed via Homebrew (brew install pixtuoid) for Linux, macOS, and WSL2, or globally via npm (npm install -g pixtuoid). After installation, launching the application with pixtuoid starts the terminal office. Agents are connected through the 'Sources' panel (s key), where users can link their agent CLIs (including Claude Code, Codex, Antigravity, and others). Pixtuoid handles the integration, allowing agents to appear in the office within seconds. A pixtuoid doctor command provides a health report for connected sources.

Supported tools include Claude Code, Codex CLI, Antigravity CLI, DeepSeek-Reasonix, and many more, with varying OS support. Configuration is managed through ~/.config/pixtuoid/config.toml, allowing customization of themes, desk capacity, and pet names. CLI flags can override configuration settings.

Technical Details

Pixtuoid operates locally with no telemetry. It processes agent events through a hook shim (writing to Unix sockets or named pipes) and JSONL transcript watching. These events are fed into a reducer that updates the office state, which is then rendered as pixel art. The core is built with five Rust crates and has no terminal dependencies. Privacy and security are emphasized, with the hook shim designed to never block agent execution and daily dependency audits.

Why it matters for builders

This tool offers a unique visual layer for managing complex AI agent workflows, transforming abstract processes into an observable, interactive environment. By consolidating agent status into a single, glanceable interface, Pixtuoid helps developers quickly identify which agents require attention, are idle, or have completed tasks, thereby improving efficiency and reducing cognitive load.

Practical impact

AI builders can install Pixtuoid via Homebrew or npm and connect their existing agent CLIs through the 'Sources' panel. Experiment with different themes and observe how agent activity, token usage, and tool selection are visually represented. Developers can also explore adding support for new agent CLIs by implementing the Source trait, contributing to the project's growing ecosystem.

Caveats and source limits

While Pixtuoid supports a wide range of agent CLIs, the source does not provide independent benchmark results comparing its monitoring capabilities against other tools. The experimental support for Windows via unsigned binaries may require additional configuration. The project is MIT licensed, and further details on contributing and architecture are available in its documentation.