MAO

Product Introduction

1. Definition: MAO is a desktop-native, GPU-rendered multi-agent orchestration (MAO) application and AI agent canvas built in Rust. It functions as a visual workspace for developers to run, coordinate, and monitor multiple AI coding agents and shell terminals simultaneously.
2. Core Value Proposition: MAO exists to solve the workflow fragmentation problem when using multiple AI coding agents like Claude Code, Codex, and Gemini CLI. It provides a unified, spatial canvas to visually manage agent terminals, explicit hand-offs, project context, and testing portals, transforming disparate command-line sessions into a coordinated, observable team.

Main Features

1. Native Multi-Agent Canvas: MAO provides a GPU-accelerated visual workspace built with the GPUI framework in Rust. It renders live terminal nodes for AI agents (Claude Code, Codex, Gemini CLI, Grok, OpenCode, Antigravity) and shell commands as movable, resizable tiles on an infinite 2D plane. This native architecture ensures high performance for complex layouts.
2. Explicit Agent Wiring & Hand-off: Users can draw visual connections between agent nodes, notes, and portals. This creates explicit dependencies and communication channels, allowing agents to delegate tasks and pass results contextually. This moves beyond manual copy-paste or shared files, formalizing multi-agent workflows.
3. Integrated Testing Portals: MAO can host live browser (Chromium) and Android device portals directly on the canvas. A coding agent wired to a portal can interact with it—clicking, typing, reading, and screenshotting—to autonomously test the web or mobile application it is building, closing the feedback loop within the same visual context.
4. Local-First & Remote Workspace Architecture: All workspace state (agent configurations, node positions, notes) is stored locally as JSON and Markdown files. This enables unlimited local workspaces. Crucially, MAO can also open a workspace on a remote Linux server via SSH, keeping the visual canvas native on your machine while agents execute on remote compute.
5. Structured Agent Presets and Roles: Users can save and assign reusable launch presets for different AI providers and shell configurations. Furthermore, reusable Markdown role definitions can be attached to agents, providing consistent context and instructions across sessions, managed via the mao CLI or the UI.

Problems Solved

1. Pain Point: Context switching and visibility loss when managing parallel AI coding sessions. Developers juggling multiple terminal tabs lose track of which agent is doing what, struggle to share context between them, and cannot see the overall workflow state.
2. Target Audience: Software engineers, DevOps professionals, and technical leads who integrate AI coding agents (Claude Code, Codex, Gemini CLI) into their daily development, prototyping, or code review workflows and need to scale beyond a single agent session.
3. Use Cases: Coordinating a team of agents where one writes backend logic (Claude Code), another writes frontend components (Codex), and a third runs tests (Gemini CLI). Running parallel refactoring sessions on different codebase modules using isolated git worktrees. Supervising an agent building a feature while it autonomously tests it in a live browser portal on the same canvas.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Unmatched visual orchestration for terminal-based AI agents; true desktop-native performance (Rust/GPUI); powerful concept of wired hand-offs and testing portals; supports both local and remote workspaces; one-time purchase pricing model with regional adjustments.
   • Cons: Requires manual setup and wiring of agents versus fully automated agentic systems; primarily a coordination layer, so users must still manage individual AI provider accounts and subscriptions; currently supports a specific set of CLI-based agents (not a generic plugin framework).

2. Key Alternatives & Differentiation:

   • vs. Conductor/Paradev: These are macOS-focused applications that automate git worktree creation for agent isolation. MAO differentiates by being cross-platform (macOS, Windows, Linux), providing a visual canvas for explicit wiring, and offering remote workspace support and testing portals, rather than being purely a local worktree manager.
   • vs. Claude Squad: Claude Squad is a terminal-based TUI that manages multiple Claude Code sessions within tmux and git worktrees. MAO offers a full desktop GUI with a spatial canvas, support for multiple AI providers beyond Claude, visual connections, and integrated portals, moving from a terminal-centric to a canvas-centric model.
   • vs. Browser-based Platforms (e.g., various cloud IDEs): Many AI agent platforms run in the cloud or a browser. MAO is a local-first, native desktop application. This ensures data privacy (workspace state stays local), lower latency for the UI, and the ability to work offline or with remote servers without sacrificing the visual interface.

Frequently Asked Questions (FAQ)

1. How does MAO handle AI provider authentication and API keys? MAO does not centrally manage keys. Each agent node uses its own provider login (e.g., claude auth for Claude Code). Users configure and authenticate each supported agent CLI locally, and MAO launches the terminal session with that context. API key rotation can be managed within saved account presets in MAO.
2. Can MAO automate workflows between agents, or is it only visual? MAO enables explicit orchestration through wiring. While it doesn't feature a fully autonomous scripting engine, wiring agents together allows them to be aware of each other's outputs and states. Hand-offs are facilitated by the user within the visual framework, providing control and oversight rather than full automation.
3. What is the difference between a MAO workspace and a project/canvas? A MAO workspace is a top-level container (represented by a local folder) that can hold multiple projects. A canvas is the infinite 2D board within a project where you place nodes. You can have multiple canvases (e.g., for different features) within a single project, which is typically a Git repository.
4. Is MAO suitable for running long-lived, persistent AI agents? MAO is designed for interactive, session-based coding workflows. It maintains terminal sessions and can recover provider sessions after a restart. However, it is not a dedicated headless agent runtime or scheduler; its strength is visual coordination and oversight during active development tasks.
5. How does the one-time purchase license work with updates? MAO is sold as a one-time purchase (not a subscription) for a major version. The listed price (e.g., $20) includes ongoing updates and support for that version line. This contrasts with subscription-based SaaS agent platforms, offering a predictable, upfront cost.