A desktop AI agent that works in your folders, on your computer
Quad Agent edits files and runs commands on your own computer. macOS, Windows and Linux are available now in public beta.
Two apps, two jobs
Quad has two desktop apps, and most people only need the first. Quad Agent is the full desktop app. It has the same chat as the web with every model in your plan, plus an Agent that reads and edits files, runs commands in a terminal and checks results in a browser, inside the folders you choose. Quad Link is a small companion that connects your computer to Quad on the web and on your phone. You can start or approve agent work from anywhere, while the work itself still runs on the computer you paired. Both apps are free to download and sign in with your Quad account. macOS, Windows and Linux are available now in public beta. Until your platform is listed, Quad runs in any browser and in the iPhone and Android apps.
What the agent does on your computer
You choose a project folder and describe the work. The agent plans, reads the relevant files, edits them, runs your commands and checks the result, showing every step in the conversation. Several details are worth knowing before you try it. Local by design. The agent works in your project folders with the terminal, browser and files on your machine. Projects and chats are stored on your computer, and prompts go to the model you select through Quad. Real tools. It uses a real terminal for commands and a separate, isolated browser for previews, with files shown in a syntax-highlighted viewer. A project can include up to 16 source folders. Several chats at once. Up to eight independent chats can run in parallel, each with its own terminal, browser and file workspace. Review what changed. A Changes tab lists staged, unstaged and untracked files with colored diffs, so you can see the project's working tree, including edits made outside the chat. Your instructions. The agent reads project-root AGENTS.md and CLAUDE.md files, and you can save custom instructions per project. Plan, then run. A Project run turns a goal into an ordered checklist of up to 16 tasks that you can edit before you start. Tasks run one after another and stop on errors, pause or app close, so nothing advances unattended. Follow-ups and commands. Queue follow-up messages while a run is in progress, and use a command palette with entries such as /review, /test and /explain. Approvals. Actions follow your approval settings. File tools use the folders you connect, but terminal commands run with your operating system account permissions: choosing a folder is not an OS sandbox. Review commands and use read access when you only need inspection.
Models, credits and privacy
Chats with Quad models use your plan's shared credits, the same ones you spend in the web app. Every model request goes through Quad's authenticated endpoint, which checks model access, reserves credits and settles the actual usage. The credit total you see is the sum of exact receipts, not an estimate. If you connect a local model instead, the agent uses no Quad credits. Quad does not use your conversations or files to train its own models. Provider API keys stay on Quad's servers, and the desktop app does not call model providers directly. Review the security page and the subprocessor list for who handles inference.
Quad Link: start work from your phone or browser
Install Quad Link on the computer you want to use, choose which folders to share, and pair it with the web app or your phone using a short, single-use code. The pairing uses encryption keys held by your devices, and Quad's relay cannot read the commands, responses or activity that pass through it. It can see that devices are connected, their IDs and the size and timing of messages. Tasks run on your computer and continue if you close the browser tab. The computer must stay on and Quad must be running, and you can pause sharing at any time.
Downloads you can verify
Today's builds are for macOS, Windows and Linux. Windows needs a 64-bit PC. Linux ships as an AppImage and a .deb installer for x64 and ARM64. Every release is described by a cryptographically signed manifest, and the download page verifies that signature in your browser before it shows a download button. It also lists the SHA-256 checksum of each installer so you can compare it after downloading. Supported direct-download builds fetch updates in the background and apply them when you restart. Package-based Linux updates can require administrator authentication.
How it compares with other coding agents
Competitor notes are from official documentation checked on September 30, 2026. Products change quickly, so confirm current features at each source. Where it runs (Quad Agent: On your computer, in folders you choose; Others, per their documentation: Codex desktop and CLI document local execution). Parallel work (Quad Agent: Up to eight chats, one per workspace; Others, per their documentation: Claude Code Desktop documents parallel sessions). Isolated worktrees (Quad Agent: Not available yet; Others, per their documentation: Codex documents local and isolated worktree execution). Undo and rewind (Quad Agent: No filesystem rewind yet; Others, per their documentation: Claude Code documents checkpoints, which do not cover shell-command changes). Diff review (Quad Agent: A Changes tab with colored diffs; Others, per their documentation: Claude Code Desktop documents diff review with line comments). Subagents (Quad Agent: Not available yet; Others, per their documentation: Codex documents specialized parallel subagents). Planning a big task (Quad Agent: Project run: an editable checklist, run in order; Others, per their documentation: Cursor introduced Projects on September 10, 2026, with delegated work). The honest summary: Quad Agent is a local agent with a good terminal, browser and file workflow and the advantage of Quad's model catalog and shared credits. Managed worktree isolation, file-level rewind, child agents and cloud fleets are not part of it today. If you need those, or deep editor integration, look at the tools above, and use Quad where its multi-model chat and local agent fit.
Try it free, no sign-up · Compare plans
Frequently asked questions
What is Quad Agent?
Quad Agent is the full Quad desktop app: Chat with every Quad model, plus an Agent that reads and edits files, runs commands in a terminal and checks results in a browser, inside the folders you choose.
Does the Quad agent run on my computer or in the cloud?
On your computer. The agent works in the project folders you choose using the terminal, browser and files on your machine. Projects and chats are stored locally, and prompts go to the model you select through Quad.
Which platforms does the desktop agent support?
macOS, Windows and Linux are available now in public beta. Windows needs a 64-bit PC, and Linux ships as an AppImage and .deb for x64 and ARM64.
What is the difference between Quad Agent and Quad Link?
Quad Agent is the full desktop app. Quad Link is a small companion that connects your computer to Quad on the web and your phone so you can start or approve agent work from anywhere. Most people want Quad Agent.
Is the desktop app free?
Both apps are free to download. Sign in with your Quad account; a free account is enough to start. Chats with Quad models use your plan's shared credits, and a local model connection uses none.
How do I know the download is genuine?
Each release is described by a cryptographically signed manifest that the download page verifies in your browser before showing a download button. The page also lists the SHA-256 checksum of every installer.
Can I run several agent tasks at once?
Yes. Up to eight independent chats can run in parallel, each with its own terminal, browser and file workspace. Project runs execute their tasks in order and stop on errors.