No. A harness is what turns an LLM into an agent: it gives the model tools, a way to call them, runs them, and hands the results back. Claude Code is kind of a harness. It is an interface (the TUI, the Code tab in the desktop app) plus a harness underneath (the Anthropic Agent SDK). Codex is the same split: the app plus the Codex App Server. T3 Code uses those harnesses. It is not a harness, and it does not work without one.
Chef = LLM. Kitchen = harness. T3 Code = the restaurant that lets you bring whatever kitchen you want. Our MCP is a cookbook we hand the kitchen, but the chef won't open it unless you tell them to :) I wrote the long version here: What is a harness?
Bring your own: Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity, Pi, or Muse Code. Provider setup
We were inspired by the Codex desktop app, Conductor, Claude Desktop, and Cursor Glass, but none met our bar. What you get:
- Bring your own sub. Claude, Codex, and whatever else you already have installed.
- Best in class remote. Run
npx t3 connecton any box and drive it from anywhere. - Best in class mobile app.
- Truly 100% open source. If we go the wrong direction, fork it.
Every harness feels nearly identical in here, so you can stop worrying about harnesses and focus on building :) Install
Yes.
Nothing. There's no way to pay us to use T3 Code. You bring your own sub. It is free and open source (MIT). You still pay your provider like you already do. T3 Connect is free too, for now. Every account can connect up to 3 machines running T3 Code. Phones and browsers that connect to them do not count. We don't want to charge for this, but if the bill gets unacceptable we may have to change course. If you hit limits or have issues, you can always fork it, self host, or use Tailscale. Install
T3 Code drives Claude through Anthropic's Agent SDK. I confirmed it's a real bug on Anthropic's side: all Agent SDK usage will drain your 5h limits faster than using Claude Code directly. To be clear: this does not affect weekly limits. We have reported it to Anthropic and are hoping for a fix soon. My post on it
Your agents run wherever T3 Code runs: your desktop, a server, a VPS. You connect to it from anywhere. I do this constantly. Less than 5% of my prompts run on my laptop. Run npx t3 connect on that machine and you get really good remote support. T3 Connect is the easy path. LAN pairing and Tailscale work too, and app.t3.codes is the web client. Remote access
Get the iOS or Android app, then connect it to your computer with T3 Connect or a pairing link. The phone is a remote. Your computer does the work and needs to stay on. Mobile setup · Remote access
T3 Connect is our built-in alternative to Tailscale for remote connections. If it keeps dropping, open Settings > Connections, find the machine, and select Details. The first part marked as failed is the cause. I will not be disappointed if you set up Tailscale as a fallback (I do this personally). T3 Connect troubleshooting
No. We work with all of the labs and T3 Code is within the TOS of Claude Code, Codex, and Google Antigravity. We worked directly with SpaceXAI to collaborate on the Grok Build and Cursor integrations. Everything else uses official ACP bindings.
If it works in T3 Code, you can trust it is safe.
I don't think T3 Code is the right place to manage a pile of accounts. If we add an auto-switcher, we are risking TOS bans for our users.
A separate work account is fine. Setup for two: Codex · Claude. For a bigger pool, manage them outside T3 Code with a self-hosted proxy like CLIProxyAPI. I walked through how I use 6 Claude subs and 3 Codex subs in this video, and wrote up the setup in this guide. Hand it to your agent and it will walk you through it. Early apologies if it gets you banned 💀 After you set up a hub, T3 Code can show its pooled limits under Usage. Connect a hub
Yes. Add another provider instance in Settings and give it its own keys and endpoint. Claude has a full OpenRouter recipe. Your subscription setup stays untouched. OpenRouter · Provider instances
No. T3 Code is a serious developer tool. Locally runnable models are not capable of meaningful engineering work.
No. Your code and prompts go from your machine to your provider (Anthropic, OpenAI, etc.), not to us. With T3 Connect, traffic rides a Cloudflare tunnel (cloudflared on your machine). Our relay sets up the link. It does not proxy your HTTP or WebSocket sessions. T3 Code does send basic product usage events (provider, model, token totals, that kind of thing). No prompts, no file contents. That is the same on Tailscale, LAN, or T3 Connect. Set T3CODE_TELEMETRY_ENABLED=false if you want it off. Product usage data
Open Settings > Storage and turn on automatic worktree cleanup. You can also run it now from there. Worktrees add up fast. One user cleared 500 GB. Storage cleanup
On iPhone, yes. You need iOS 26 or later. Tap the mic in the composer, talk, check the text, then send. Web, desktop, and Android do not have it yet. Voice input on iPhone
Yes, all three. Desktop app, CLI, the works. One catch: there is no t3 binary for Intel Macs, so use the desktop app there. Windows folks can also run agents inside WSL. Install
Almost always, the CLI is not installed or not on the PATH of the machine that runs the server. (Not your phone. The computer.) Log in on that machine first, like codex login or claude auth login. Still missing? Point Binary path at it in provider settings. Version managers love to cause this one. Provider setup