offstage logo

offstage

Freemium
Visit

A macOS tool that gives coding agents a separate desktop for GUI work without taking over the user’s screen, keyboard, or mouse.

What is offstage?

offstage gives a coding agent a second, logged-in macOS user account on the same Mac. GUI work—such as simulators, Xcode UI tests, headed browser sessions, and built applications—runs in that account’s desktop while the primary user keeps their own screen, keyboard, mouse, and active windows. The tool is intended to keep agent-driven interface work from taking focus on the user’s console.

It routes commands according to whether they need a graphical environment. Headless tests run where they already run, while supported GUI commands use the helper account. offstage is a second-user workflow rather than a virtual machine, so both accounts share the Mac’s CPU, memory, and disk.

What can offstage do?

Separate macOS desktop

Runs GUI commands in a logged-in helper account called computeruse, with its own desktop, window server, and input context. The primary user’s display and input remain separate.

Command routing

Reads commands before execution and selects a macOS session, a Linux container with a virtual display when Docker is installed, the current environment for headless work, or no lane for refused installers and disk images.

Agent integrations

Works with Claude Code, Codex, and opencode through MCP. Claude Code can also use offstage as a plugin, and the CLI provides `offstage route` and `offstage run`.

Interactive session controls

Supports an app-testing loop with session launch, screenshots, input, and quit operations. Screenshots use point coordinates, and the launch operation waits for the app to register.

Read-only project sharing

Lets the helper account access a selected project folder in read-only mode when normal account permissions prevent a run. Sharing can later be removed.

Diagnostics and safety rules

The `offstage doctor` command reports which execution lanes work and how to fix them. Refused installers are not overridden, and input is rejected if the helper session is the active console session.

Use Cases

“Run simulator and Xcode UI tests”

A coding agent can run simulators, `xcodebuild test`, `xcrun simctl`, XCUITests, and related app commands in the helper account instead of opening windows on the developer’s desktop.

“Test a built Mac app interactively”

The agent can launch a registered app, inspect session screenshots, send input, and capture another screenshot to verify the result without taking over the user’s active session.

“Run headed browser workflows”

Headed Playwright, Puppeteer, Cypress, and WebDriver workflows can use a separate macOS desktop; supported headed browser work may also use a Linux virtual display when Docker is installed.

“Keep routine test runs local”

Headless Playwright or Puppeteer and ordinary commands such as npm test or vitest remain in their existing environment, avoiding the overhead of routing work that does not open a window.

Frequently Asked Questions

Is offstage a virtual machine?

No. It uses a second ordinary macOS user account on the existing Mac, with no guest operating system or VM to boot. The product site says the setup takes about 3 GB of disk.

What is required for initial setup?

The site lists Apple Silicon Macs and Node.js 20 or newer. Setup includes installing the CLI, running `sudo offstage session setup --create`, and switching to the helper account once so its daemon starts. The setup needs administrator access and may require enabling Screen Recording and Accessibility permissions.

Which coding agents are supported?

Claude Code, Codex, and opencode are supported through MCP. Claude Code is also supported as a plugin. The CLI can be used when MCP is unavailable.

Can the helper account access my project files?

It is an ordinary separate user and cannot write to the primary user’s files. If it cannot read a project, `offstage session share <dir>` grants read-only access to one folder at a time; the share can be removed afterward.

Which commands does offstage refuse?

Installer and disk-image operations, including `.pkg`, `.dmg`, and `hdiutil`, are refused. The site also states that headless commands should run in place rather than being wrapped unnecessarily.

Quick Facts

Category
Developer tool for agent-driven GUI work
Platform
macOS on Apple Silicon
Runtime requirement
Node.js 20 or newer
Supported agents
Claude Code, Codex, and opencode
Execution model
Second logged-in macOS account on the same Mac
License and price
Free; MIT licensed

offstage Traffic Analysis

Traffic data is for reference only.

Domain Rating
0

offstage Alternatives

Firetower logo

Firetower

usefiretower.com

Firetower is an open-source, self-hosted control plane for running coding agents on servers you own. It lets developers start agent sessions from repositories, monitor them from desktop or mobile clients, review changes, and ship branches without keeping the initiating device online.

Kody logo

Kody

kody.codes

Kody is a software platform that gives connected AI agents shared memory, credentials, reusable packages, integrations, and scheduled execution. It helps developers carry work between agents such as Cursor, Claude, ChatGPT, and Codex without rebuilding the same automation each time.

treg.to logo

treg.to

treg.to

treg.to is an API gateway and MCP server for calling tools and data providers through one endpoint and credential. It helps agents and developers use cataloged SEO, enrichment, advertising, social, scraping, and other data tools without managing separate provider keys for every metered call.

Zoku logo

Zoku

heyzoku.com

Zoku is a native macOS app for running and coordinating multiple coding-agent CLIs on one canvas. It lets developers spawn, redirect, monitor, and interrupt named agents by voice while using the subscriptions and command-line tools already installed on their Mac.

OrgComputers logo

OrgComputers

www.orgcomputers.com

A shared workspace that gives AI coding agents common project context, memory, and tools.

Dezycro logo

Dezycro

app.dezycro.ai

Dezycro is an operations layer for AI coding agents that provides product context, steers work, verifies behavior against requirements, and records decisions. It is designed for developers, product teams, and organizations that want to use agents without giving Dezycro access to source code.