What Is T3 Code? A Free Control Plane for All Your AI Coding Agents

If you use several AI coding agents at once, you know the routine: a pile of terminal windows, one agent and one project in each, and the job of remembering which agent is doing what. T3 Code was built to bring all of that into one place.

T3 Code is a free app, open source under the MIT license, built by Theo Browne and Julius Marminge at Ping Labs. Its homepage, t3.codes, describes it as a “control plane for coding agents”: a single dashboard for the agents you already use, such as Codex, Claude Code and Cursor.

This post explains what T3 Code is, how it differs from the other T3 products, which agents it supports, its key features, how it works under the hood and how to get started.

What Is T3 Code?

T3 Code is a graphical interface for running and orchestrating AI coding agents. It has no AI model of its own and does not resell tokens. Instead, it drives each vendor's official tools on your machine, using the accounts and subscriptions you already have.

You can use T3 Code on desktop (macOS, Windows, Linux), on the web at app.t3.codes, on iOS and Android, or through the t3 command in your terminal. Asked “what are you selling me?”, the project README answers simply: “Nothing.”

How Is T3 Code Different From T3 Stack and T3 Chat?

Theo Browne has three products with the T3 name, which are easy to mix up:

• T3 Stack (create-t3-app): a starter for web projects with Next.js, tRPC, Prisma and Tailwind.
• T3 Chat (t3.chat): a chat app for multiple AI models.
• T3 Code (t3.codes): an app for orchestrating coding agents, and the subject of this post.

When Did T3 Code Launch?

According to the release history on GitHub, the key milestones are:

• February 8, 2026: the pingdotgg/t3code repo was created.
• March 2, 2026: the first alpha build.
• March 7, 2026: v0.0.2, the public launch window.
• October 15, 2026: the stable v0.1.0 release, called Orchestrator V2.
• October 22, 2026: v0.1.2, the latest release at the time of writing.

The project ships fast, with nightly builds several times a day. As of October 24, 2026, the repo had about 29,600 stars and 7,900 forks on GitHub.

Which Coding Agents Does T3 Code Support?

According to the README, T3 Code supports Codex, Claude Code, Cursor, Grok Build, OpenCode and Google Antigravity. Version 0.1.0 added Pi, Muse Code, OpenCode 2 and agents from the ACP Registry such as Devin, Cline, Kimi and Droid.

What stands out is how T3 Code connects. It uses each vendor's official programming interface instead of reading terminal output: Codex through the Codex app-server, Claude through the Claude Agent SDK, Cursor through its official SDK, Grok through the Agent Client Protocol and OpenCode through the OpenCode SDK. Events from every agent are normalized into one shared model, so you work with all of them the same way.

To sign in, you log in to each tool as usual, for example codex login or claude auth login, and T3 Code reuses that session. Codex also offers a Connect with ChatGPT flow that uses your ChatGPT plan directly. If you prefer, API keys, OpenRouter, Bedrock and Vertex work through environment variables.

Key Features of T3 Code

T3 Code is built around a git workflow that lets you run many agents in parallel without their code colliding:

• A branch per thread: every agent thread works on its own branch or worktree.
• In-app diffs: review changes before you push.
• One-click PRs: commit, push and open a pull request with an auto-generated title and description, on GitHub, GitLab, Bitbucket and Azure DevOps.
• Four permission modes: Supervised, Auto-accept edits, Auto and Full access, so you decide how much the agent can do on its own.
• Checkpoints: roll back to any message in the conversation.
• Approve from any device: answer agent approval requests from desktop, browser or phone.
• Usage stats: see your usage converted into an API-equivalent cost.

What Is New in v0.1.0 (Orchestrator V2)?

According to the v0.1.0 release notes, the first stable release brings deeper multi-agent orchestration:

• Cross-agent delegation: the delegate_task tool lets one agent spawn child agents on any provider and wait for the result, for example Claude planning and Codex implementing.
• Thread forks: split one thread into several directions and merge them back.
• T3 Code MCP: lets agents create, message, wait on and stop other threads.
• Switching providers mid-thread.
• Scheduled tasks and auto-resume when usage limits reset.

For remote work you have three options: T3 Connect (a relay run by Ping Labs), direct pairing over LAN or Tailscale with a QR code, or SSH.

How Does T3 Code Work Under the Hood?

According to the architecture docs, the central principle is that the machine holding the workspace executes everything. Each machine runs its own server, and the web, desktop and mobile clients only control that server over an authenticated connection. The desktop app bundles this server.

State is stored with event sourcing in SQLite under ~/.t3/userdata, and checkpoints are kept as hidden git refs. Code, prompts and responses stay on your machine. The T3 Connect relay only helps establish the connection and does not carry your session content.

T3 Code sends anonymous usage data such as provider, model and token totals, but not prompts, responses or file contents. According to the telemetry docs, you can turn it off with the environment variable T3CODE_TELEMETRY_ENABLED=false.

How to Install and Get Started With T3 Code

You can get going in three steps:

• Step 1: install and log in to at least one coding agent, for example codex login or claude auth login.
• Step 2: install T3 Code with npx t3@latest (requires Node.js 22.16 or later), or through Homebrew (brew install --cask t3-code), winget (winget install T3Tools.T3Code), scoop, a .deb package or the AUR.
• Step 3: run t3 to start the server and open the app.

When you start out, pick Supervised mode and try a small task. Then review the diff, try rolling back with a checkpoint, and only then give the agent more permissions or turn on remote access. You can also run t3 service install to keep T3 Code running as a background service.

Who Should Use T3 Code?

T3 Code fits developers who already use one or more coding agents and want a cleaner way to manage them:

• You run several projects or agents in parallel and want to track them all in one place.
• You want to combine the strengths of different models in the same piece of work.
• You work on Linux or Windows, or want to approve agent work from your phone.
• You are a freelancer juggling projects for several clients at once.

If you want your team to master AI coding tools like T3 Code, Codex and Claude Code, Growthtrait's AI training program can help you build the right workflow. Contact us to talk it through.