@theo
Had 26 threads running in T3 Code across his fleet of machines, then clicked "update all" for T3 Code, Codex, Claude Code and Muse, trusting T3 Code to restore the threads.
Repo of the Week · Week 45, 2026
T3 Code: run your coding agents from one app, on any device
A free app that runs Claude Code, Codex and other coding agents on your computer and lets you steer them from your phone, browser or desktop.
What’s a repo?A repo is a project’s code folder on GitHub, free to download. pingdotgg/t3code on GitHub
⭐ 38.8k starsFree · MITClaude Code, Codex, Cursor and more
An open-source app to control Claude Code, Codex, Cursor and other agents on your computer from desktop, web or phone.
- Stars
- 38.8k
- Forks
- 9k
- npm / week
- 34.8k
- Latest
- v0.1.6
💡
What it is
You keep the coding agents and subscriptions you already have. T3 Code puts them in one window, each task in its own thread, and lets you check on them or answer their questions from your phone while your computer does the work.
🧑💻
Who it’s for
People who already use one or more coding agents (Claude Code, Codex, Cursor and others) and want one place to run several tasks at once, or to keep an eye on them away from the desk.
🧰
What you need
At least one installed and signed-in coding agent (for example Claude Code or Codex) on the computer that will do the work, running Windows, macOS or Linux. For phone access, that computer must stay on and reachable through T3 Connect, your local network or Tailscale.
Why it’s trending
Created Feb 8, 2026, it now has 38,752 stars and 8,957 forks (GitHub API, Nov 2, 2026). GitHub Trending shows 6,926 new stars this week. On X, 1,765 original posts mentioned T3 Code in the past 7 days; the 89 we read came from 70 different people. Here’s what people are sharing:
- Official
Day 31: T3 Code has /btw and side chats now, and they work with any model or harness: 2,480 likes, 40 reposts, 225 replies.
@theo · Oct 31, 2026 ↗ - How-to
Day 27: hit a bug? Command-K "triage" clones the T3 Code source for your exact version, gives your agent a playbook to debug and fix it, and cuts an issue if it's their fault: 525 likes, 165,138 views.
@theo · Oct 28, 2026 ↗ - Trending
T3 Code gets more engagement on GitHub than VS Code: 804 likes.
@theo · Oct 29, 2026 ↗ - Trending
pingdotgg/t3code is the #5 most engaged repo, with anomalyco/opencode at #2.
@MeetPatelTech · Oct 28, 2026 ↗ - Trending
GitHub Trending today, #1: pingdotgg/t3code, +3,100 stars today, 37,713 total.
@Svenu1s5 · Nov 2, 2026 ↗ - Reaction
Compares agent workspaces that isolate each task in a Git worktree: Orca, T3 Code and Tern as GUI control surfaces against terminal-native Herdr and Rex (in Chinese).
@jiangkoumo_ · Oct 28, 2026 ↗ - Caution
Theo says an in-app note recommending a GitHub star after you've sent 5 prompts immediately 10x'd the stars per day: 1,148 likes.
@theo · Nov 2, 2026 ↗ - Concern
A verification task used 40% of his Claude limit in T3 Code, against 8 to 10% for the same task in the Claude CLI.
@amandotdev · Nov 1, 2026 ↗ - Concern
How do you bring a decision from a side chat back into the main run without copying the whole detour?
@progerallworld · Oct 31, 2026 ↗
What it does
An open-source "agent harness control surface" built by Theo (@theo) and the pingdotgg team. It controls the coding agents already installed on your computer: Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity, Pi, Muse Code and fx. A local server runs next to your agents, and a desktop app, a web app and iOS and Android apps connect to it to start threads, review diffs, answer approvals and follow progress. It uses your existing subscriptions. It is not a harness itself and does not work without one.
One app for many agents
Claude Code, Codex, Cursor, Grok Build, OpenCode, Antigravity, Pi, Muse Code and fx, using the subscriptions already set up on the machine
Phone, web and desktop
iOS and Android apps, the app.t3.codes web client and an Electron desktop app all connect to the same server on your computer
Remote access
t3 up sets up T3 Connect (a Cloudflare tunnel), local network pairing or Tailscale; code and prompts go to your provider, not to T3 servers
Threads and worktrees
Each task runs in its own thread, optionally in its own Git worktree, with a diff pane to review the changes
Permission modes
Supervised, Auto-accept edits, Auto or Full access per thread; an administrator can cap them with a managed policy
Side chats
Start a message with /side or /btw to ask about a thread without adding to it; Bring back hands what you learned to the thread (v0.1.6)
MCP for outside agents
Claude Code, Codex, ChatGPT and other MCP clients can read and start threads through the environment's /mcp address, read-only by default
What people built with it
@ryanrhughes
Says an AppSignal error could start an agent investigation through a T3 Code webhook with his laptop closed, and that features like this are why he tells everyone to try it.
@KoiLx0
Codes on all his servers straight from his phone with T3 Code plus Tailscale.
@alvaris9240
His team chose T3 Code as the foundation for user-facing AI automation features (AI Daily Standup, Night Shift), citing its interface, remote access and multi-agent support.
@AvidNetizen
T3 Code has no native GitHub Copilot support, so he has Claude call the Copilot CLI through T3 Code for his PR reviews.
@dronathon
Ran DeepSeek Harness using T3 Code for Claude Code and Codex, with no forks and T3's native interface.
@preferedev
Forked the MIT-licensed code to build Hybrid Bot: describe an idea and it sends a dev agent to build it, reusing T3 Code's projects, threads, diffs and checkpoints.
How to use it
Set up a coding agent first
T3 Code drives agents you already have. On the computer that will do the work, install at least one and sign in. For Claude Code:
claude auth loginOr for Codex:
codex loginCursor uses
agent login, Grok Buildgrok loginand OpenCodeopencode auth login.Install T3 Code
On macOS or Linux, run:
curl -fsSL https://t3.codes/install.sh | shOn Windows, in PowerShell:
irm https://t3.codes/install.ps1 | iexPrefer a desktop app? Download it from GitHub Releases, or run
brew install --cask t3-code(macOS) orwinget install T3Tools.T3Code(Windows).Start it
t3This starts the server and opens the local web app. If your shell says
command not found, add~/.local/binto your PATH; the installer prints the line to add.Start a thread
Pick your project and a provider, then type your task. Choose a permission mode in the composer: Supervised asks before commands and file changes.
New threads start in Full access unless you change it in Settings → Threads → New threads → Permissions.
Connect your phone
Install the iOS or Android app. On the computer, run:
t3 upChoose T3 Connect and sign in, then sign in to the same account on the phone. The phone is a remote: your computer does the work and must stay on.
Prerequisites
- An installed and signed-in provider on the computer that runs T3 Code, such as Claude Code (claude auth login), Codex (codex login), Cursor CLI (agent login), Grok Build (grok login) or OpenCode (opencode auth login) (README)
- Windows, macOS or Linux. There is no t3 command-line binary for Intel Macs; use the desktop app there (install docs)
- Node.js only if you try it once with npx t3@latest instead of installing (install docs)
- For phone or remote use: the computer running T3 Code stays on and reachable (remote access docs)
- No payment: T3 Code is free, and T3 Connect is free for up to 3 machines (FAQ)
Try it once without installing
Needs Node.js for npx; nothing is installed.
npx t3@latestKeep it running in the background
Runs T3 Code as a background service on macOS and Linux (systemd user service on Linux). t3 status shows how your other devices reach it. Windows has no background service; use t3 serve.
t3 service install
t3 statusPair over your network instead of T3 Connect
Accepts connections from your local network and tailnet. Scan the QR code or paste the pairing URL into Add environment. On that network, the pairing code is all that keeps other devices out.
t3 up --lan
t3 pairLet outside agents use it over MCP
Copy the MCP URL from Settings → Environments. Agents get Read only access by default; choose Supervised through Full access to let them start and message threads. For Codex: codex mcp add t3 --url https://<environment-address>/mcp.
claude mcp add --transport http t3 https://<environment-address>/mcp
claude mcp login t3Script threads on the host
--wait prints the agent's reply. Add --worktree <base-branch> to t3 thread start to work in a new worktree.
id=$(t3 thread start . "Find why the tests are slow")
t3 thread send "$id" "Now fix it" --waitAsk on the side
A side chat starts with what the thread knows and doesn't add to it. Choose Bring back while the thread is idle to hand it what you learned.
Prompt
/btw why did you pick this library?Triage a T3 Code bug with your own agent
Or choose Triage an issue in the command palette. Your agent reads the logs and the T3 Code source for your version, and asks before it applies a fix or posts anything.
npx t3 triageStay up to date
Set T3CODE_CHANNEL=nightly when you install to follow nightly builds, or T3CODE_VERSION to pin an exact version.
t3 updateGotchas
- Claude's 5-hour limit drains faster through T3 Code. Its FAQ says all Anthropic Agent SDK usage does this compared with Claude Code directly (a bug reported to Anthropic); weekly limits are not affected.
- The phone is only a remote. The computer running T3 Code must stay on, awake and reachable.
- New threads default to Full access: agents run commands and edit files without asking.
- No local models. The FAQ says T3 Code does not support them.
- Worktrees add up fast. Turn on automatic cleanup in Settings → Storage.
- It sends basic usage events (provider, model, token totals; no prompts or file contents). Set
T3CODE_TELEMETRY_ENABLED=falseto turn them off.
Before you start
- Most of the likes come from its creator. In our sample, posts by @theo hold 13,294 of 13,386 likes. Other people posted a lot (1,726 posts in the week), but most of those posts got little engagement.
- Star growth was nudged. On Nov 2, 2026 Theo said an in-app note asking for a star after 5 prompts 10x'd the stars per day, so read the star count with that in mind.
- The README says the project is very early and to expect bugs. Stable releases come every few days and nightly builds several times a day.
- Contributions are mostly not accepted yet. Small fixes may be considered; big features will not be.
- New threads default to Full access, so agents can run commands and change files without asking. Pick Supervised if you want to approve each step.
- Driving Claude through T3 Code uses up the 5-hour limit faster than Claude Code directly (FAQ; a bug reported to Anthropic).
FAQ
What is T3 Code?
An open-source "agent harness control surface" built by Theo (@theo) and the pingdotgg team. It controls the coding agents already installed on your computer: Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity, Pi, Muse Code and fx. A local server runs next to your agents, and a desktop app, a web app and iOS and Android apps connect to it to start threads, review diffs, answer approvals and follow progress. It uses your existing subscriptions. It is not a harness itself and does not work without one.
Is it free?
Yes. pingdotgg/t3code is MIT-licensed and its FAQ says there is no way to pay for T3 Code. You still pay your agent provider as usual. T3 Connect is free for up to 3 machines; a paid T3 Chat plan raises that to 10 (Pro) or 100 (Premier).
Can I use my Claude Code or Codex subscription?
Yes. It drives the agents already installed and signed in on your computer. The FAQ says this is within the terms of Claude Code, Codex and Google Antigravity, and that other providers use official ACP bindings.
Why does my Claude limit run out faster in T3 Code?
People keep asking this on X. The FAQ says it is a bug on Anthropic's side: Agent SDK usage, which T3 Code uses to drive Claude, drains 5-hour limits faster than Claude Code directly. Weekly limits are not affected, and the bug has been reported to Anthropic.
Does my code go to T3's servers?
No, per the FAQ. Code and prompts go from your machine to your provider. T3 Connect uses a Cloudflare tunnel, and its relay only sets up the link. Basic usage events (no prompts or file contents) are sent unless you set T3CODE_TELEMETRY_ENABLED=false.
Does it work on my phone?
Yes, through the iOS and Android apps, connected with T3 Connect or a pairing link. The phone is a remote: your computer does the work and has to stay on. Voice input works on iPhone only (iOS 26 or later).
Is it a harness like Claude Code?
No. The FAQ says T3 Code uses harnesses such as Claude Code or Codex and does not work without one.
Does it support local models?
No. The FAQ says it does not support locally run models.
How is the Repo of the Week picked?
We feature the GitHub repo most talked about on X that week in AI and vibe coding, counted by distinct authors and engagement, and cross-checked against GitHub Trending. Stats come from the GitHub API and are dated. T3 Code was checked 2026-11-02.
All weeks
Every pick stays here with its own page. Newest on top.
- Week 45, 2026This week
T3 Code: run your coding agents from one app, on any device
A free app that runs Claude Code, Codex and other coding agents on your computer and lets you steer them from your phone, browser or desktop.
- Week 44, 2026Last week
Security Audit Skill: Cloudflare's security audit for your coding agent
A free add-on that makes your AI coding agent check your code for security holes, with a second agent double-checking every finding.
- Week 43, 20262 weeks ago
HyperFrames: write HTML, render video
A free tool that lets your AI coding agent make real videos by writing a web page.
- Week 42, 20263 weeks ago
REA: Reverse Engineer Anything
A free tool that lets your AI coding agent take apart any app to see how it works.