T3 Code Review

T3 Code is described as an agent harness control surface: it controls the agents on your machine with a mobile app (iOS and Android), a web app, and an Electron-based desktop app. It works with your subscriptions to Claude Code, Codex, Cursor, Grok Build, OpenCode, and Google Antigravity, and if those are set up on your computer, T3 Code can drive them. It does not sell anything: the README answers the question what are you selling me with nothing, and says the team built it because they wanted a fast, remote-ready, truly open tool and none of the existing options (the Codex desktop app, Conductor, Claude Desktop, Cursor Glass) met their bar. If the project goes in the wrong direction, they want you to be able to fork it. The documentation folder covers install and first run, permission modes, keyboard shortcuts, project settings, remote access from a phone or another machine, connecting Claude Code, Codex, and ChatGPT over MCP, running as a background service, and source control integrations, but there is no docs site yet.

You must first install and authenticate at least one provider (for example install the Codex CLI and run codex login, or Claude Code and claude auth login). Then install with curl and sh from t3.codes (or irm with PowerShell on Windows), run t3 to start the server and open the web app, t3 service install to keep it running, and t3 update for new releases; npx t3@latest tries it without installing. Desktop packages exist for winget, scoop, Homebrew, .deb, and the AUR. Contributing needs the vp tool from Vite+. We did not run it.

A single interface for starting and watching agent sessions across providers, usable from a phone when your computer is running the server. The README's own warning that the project is very early applies to the experience. We did not test it.

• Controls Claude Code, Codex, Cursor, Grok Build, OpenCode, and Antigravity
• Web, Electron desktop, iOS, and Android apps
• Remote access from a phone or another machine
• Runs as a background service
• MCP connection for external agents
• MIT license

• Single control surface across several competing agents
• Mobile and remote access for long-running agent sessions
• Fully open under MIT, built to be forkable
• Rapid traction: about 27,900 stars

• Authors say they are very early and expect bugs
• Mostly not accepting contributions
• No documentation site yet, only a docs folder
• Requires separately installed and authenticated agent CLIs

T3 Code is free under the MIT license. It uses the subscriptions or API access you already have for each agent.

Vendor apps such as the Codex desktop app or Claude Desktop control a single provider's agent. T3 Code is vendor-neutral and open source, so you can switch between agents in one interface, with the early-stage caveats the authors describe.

• Developers who use multiple coding agents
• People who want to supervise agent sessions from a phone
• Teams that prefer open-source tooling

• Anyone who needs a stable, polished product today
• Developers who want to contribute large features right now
• Users without any supported agent installed