Introduction

Agent Grid turns "talking to Claude" or "running Codex" into something you do on a 2D canvas with as many parallel agents as you need — side by side, with the terminals, browsers, code editor, source control, and notes that support them.

What it is

Agent Grid is a desktop app (macOS, Windows, Linux) where every AI agent, terminal, browser, file you're reading, and note lives as a pane on an infinite canvas. You spawn panes, drag them around, and let them work in parallel.

The mental model

Four nouns. Learn them and the rest of the docs make sense.

• Canvas — the infinite, pan-and-zoom 2D surface. One per space.
• Space — a tab in the title bar. Each space binds a single project folder (the working directory for new agents and terminals) and owns its own pane list.
• Pane — a unit on the canvas: a Claude Code instance, a Codex instance, an Antigravity (agy) instance, a terminal, a code editor, source control, a browser, a note, an image, a video.
• Agent — an AI instance running inside a Claude, Codex, or Antigravity pane. A Claude pane that drives other panes via MCP is called a master (or orchestrator); the agents it spawns are workers.

Who it's for

• Developers running more than one Claude Code or Codex session at a time who keep losing track in a stack of terminal tabs.
• Teams adopting orchestrator/worker patterns — one Claude pane spawning specialized workers (builder, qa, security, …) and watching them in parallel.
• Anyone who wants their agents on the same surface as the terminals, browsers, and notes that support them.

Next steps

• System requirements — confirm your machine is supported (macOS ships for Apple Silicon and Intel).
• Installation — download and install per platform.
• CLI prerequisites — install and authenticate claude, codex, and agy.
• Quickstart — create a space, pick a project folder, spawn your first Claude agent.

Once you're set up, the Core Concepts section goes deeper on the canvas, spaces, panes, and the master/worker model.