What Is T3 Code?
T3 Code calls itself an "agent harness control surface", which is a grand way of saying one thing: it is the remote control for the AI coding agents running on your machine. You open an app on your iPhone, your Android phone, the web or a desktop app, and you drive the agents that are already set up on your computer, from wherever you are.
It works with the subscriptions you already have. Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity and fx are all supported, and if a provider is installed and authenticated on your machine, T3 Code can control it. There is no new subscription to buy and no new model to pay for. The README asks the question itself, "Wait, what are you selling me?", and answers it in one word: "Nothing."
The project is written in TypeScript, licensed under MIT, and it had 45,475 stars when we checked on November 4, 2026, after gaining roughly 12,500 stars in the past week. The team says they built it because they wanted the best possible development experience with agents, inspired by the Codex desktop app, Conductor, Claude Desktop and Cursor Glass, but none of those met their bar for something performant, remote-ready and truly open. If the project ever goes the wrong direction, they want you to have everything you need to fork it and build the editor you want.
One App for Every Agent You Already Pay For
The practical appeal is consolidation. If you use Claude Code for one project, Codex for another and Cursor somewhere else, each normally has its own interface and its own way of being driven. T3 Code puts them behind one control surface, so you check on runs, steer agents and keep working from your phone instead of being chained to the machine the agents run on.
Setup follows the provider, not T3 Code. Install and authenticate at least one provider before use: Codex with the Codex CLI and codex login, Claude with Claude Code and claude auth login, Cursor with the Cursor CLI and agent login, Grok Build with grok login, OpenCode with opencode auth login, and fx with fx login on the machine running the T3 Code server. Antigravity is the exception: enable it in Settings, then use Install Antigravity and Sign in with Google, with no CLI required.
There is also a docs folder in the repository covering the details that matter once you are running: permission modes, keyboard shortcuts, project settings, appearance preferences, remote access from a phone or another machine, connecting agents over MCP, source control integrations, multiple accounts per provider, and running T3 Code as a background service. There is no docs site yet, so the repository docs are the documentation.
Install It in One Command
The fastest route is the install script, then the t3 command to start the server and open the local web app:
curl -fsSL https://t3.codes/install.sh | shOn Windows, the PowerShell equivalent is irm https://t3.codes/install.ps1 | iex. Once installed, t3 service install keeps the server running in the background, t3 update moves to a newer release, and t3 --help has the full reference. To try it once without installing anything, run npx t3@latest instead.
Prefer a desktop app? The latest release is on GitHub Releases, and the package managers are covered: winget install T3Tools.T3Code on Windows, scoop install extras/t3code, brew install --cask t3-code on macOS, a .deb for Debian and Ubuntu, t3code-bin or t3code-nightly-bin from the AUR on Arch, and a nixpkgs package that includes both the desktop app and the t3 command. The mobile apps are on the App Store and Google Play, and the web app lives at app.t3.codes.
What to Watch Out For
The README is blunt about the project's age, and you should take it seriously: "We are very very early in this project. Expect bugs." A control surface for your coding agents sits in a privileged position, so running an early project here deserves more caution than installing a text tool. Keep it on a machine you can afford to babysit, and read the permission modes doc before you hand an agent a long leash. Supervising from a phone also makes it easy to approve work you have only skimmed, so read the actual changes at a desk before you merge: a text diff shows every line an agent rewrote, and the AI Code Detector flags the placeholder scaffolding and debug leftovers agents tend to leave behind.
Contributions are mostly closed for now. The team says small fixes may be considered, but big features will not be, so if you adopt T3 Code hoping to shape its roadmap, expect to wait. Building from source is also a heavier lift than usual: it uses Vite+ and needs the global vp command-line tool, with full instructions in the repository's contributing guide.
None of that cancels the core idea. If you already pay for two or three agent subscriptions and you want to supervise them away from your desk, a single open source remote control with no subscription of its own is a genuinely useful shape. Just go in knowing it is early, and let the phone app earn your trust before you give it the keys to anything important.
