T3 Code documentation, in plain English
A community reference for T3 Code, the open-source GUI for Codex, Claude Code, Cursor and OpenCode. Install guides, remote access, worktrees and real fixes.
T3 Code is an open-source, MIT-licensed web GUI for coding agents. Instead of running Codex or Claude Code in a terminal tab, you get a browser (or desktop) interface with threads, a diff viewer, git worktree management and one-click pull requests — all driving the same agent CLIs you already have installed.
This site is a community reference. The project moves fast — nightly builds ship most days — so the goal here is a stable, searchable place for the things the official docs cover briefly or not at all: the exact install command per platform, what actually happens on a remote box, and the error messages that send people to GitHub issues.
Jump straight in#
- Quickstart — from zero to a working agent thread
- Installation — npx, Windows, macOS, Linux
- Remote access — run the server on one machine, use it from another
- Troubleshooting — the seven errors that account for most support traffic
- Alternatives — how it stacks up against Conductor, Cursor and the Codex app
- News — what changed in recent builds
What T3 Code actually is#
It helps to be precise, because the name causes confusion. T3 Code is not an AI model, and it is not Create T3 App (the Next.js/tRPC/Prisma starter from the same people). It is a client — a front end that talks to coding agents you already pay for.
| It is | It is not |
|---|---|
| A GUI wrapper around agent CLIs | An AI model or an API provider |
| Bring-your-own-key / your own subscription | A new subscription to buy |
| Open source, MIT licensed | Closed or paid tiers |
| A thread + git workflow layer | A full IDE replacement |
| Web app + Electron desktop app | A VS Code extension |
Under the hood, your browser holds a WebSocket connection to a small Node server on your machine. That server launches the provider CLI as a subprocess and speaks JSON-RPC to it over stdio, then streams structured events back to the UI. Nothing is proxied through a T3 server — see architecture for the full path.
Supported providers#
Four agents are supported today. You install and authenticate each one yourself; T3 Code just drives it.
| Provider | Auth command | Notes |
|---|---|---|
| Codex | codex login | The original and best-tested integration |
| Claude Code | claude auth login | Works with a Claude subscription or API key |
| Cursor | cursor-agent login | Uses the Cursor agent CLI, not the IDE |
| OpenCode | opencode auth login | Open-source agent, broad model support |
Asking about Gemini, Ollama, OpenRouter or local models? That page covers what is and isn't wired up, and the honest answer for each.
Why the git integration is the actual selling point#
Most people arrive for "a nicer chat window" and stay for the git plumbing. Each thread can be mapped to its own git worktree, so three agents can work on three branches in three directories at once without stepping on each other. When one finishes, the diff viewer and a single control cluster take you from review to commit to pushed pull request.
That is the workflow difference versus a terminal: you are not the one running git worktree add, remembering which tab is on which branch, or writing the PR body.
Cost#
T3 Code itself is free and MIT licensed — there is no paid tier and no account to create. You pay whatever your underlying provider costs: an OpenAI or Anthropic subscription, or per-token API usage. The FAQ covers the details, including which subscriptions carry over.
About this site#
Every page here records the date it was last checked against the pingdotgg/t3code repository. Where a page describes behaviour that comes from a specific GitHub issue or release, it links to it, so you can confirm whether it's still true.
For anything canonical — API guarantees, security posture, roadmap — go to the official documentation, the repository, or the Discord. This site is not affiliated with Ping Labs or t3.gg.
FAQ#
Is T3 Code free?#
Yes. T3 Code is open source under the MIT license with no paid tier, no subscription and no account. You pay only for the coding agent you connect it to — your existing OpenAI, Anthropic or Cursor subscription, or per-token API usage on your own key.
What is T3 Code used for?#
It gives coding agents like Codex and Claude Code a graphical workspace instead of a terminal. You get persistent chat threads per project, a diff viewer, git worktree isolation so multiple agents can run in parallel, and a one-click commit-push-pull-request flow.
Is T3 Code the same as Create T3 App?#
No. Create T3 App is a scaffolding CLI for full-stack TypeScript projects. T3 Code is a desktop and web GUI for AI coding agents. They share the T3 name and originate from the same community, but they are unrelated tools.
Does T3 Code need Codex installed?#
You need at least one supported provider CLI installed and authenticated. Codex is the most mature integration, but Claude Code, Cursor and OpenCode also work. T3 Code launches the CLI as a subprocess, so it cannot authenticate on your behalf.
Is T3 Code production ready?#
Not yet. The project describes itself as very early stage, ships nightly builds most days, and warns about bugs and breaking changes. It is stable enough for daily development work, but pin a version if you need reproducibility.