T3 Code: The Open-Source Tool With 17,700+ GitHub Stars That Lets You Control AI Coding Agents From Anywhere

Every developer using AI coding agents has hit the same wall: you start a long-running task on your workstation, then you're chained to that chair. Step away and you lose visibility. Switch devices and you lose control. T3 Code changes that equation entirely.

Built by the team behind ping.gg — the developer tooling company known for uploadthing and create-t3-app — T3 Code is an open-source "agent harness control surface" that wraps every major AI coding agent CLI in a unified, remote-ready interface. And it just hit 17,755 stars on GitHub with 4,000+ forks, making it one of the fastest-growing developer tools of 2026.

What Is T3 Code, Exactly?

T3 Code isn't an AI coding agent itself. It's the control layer that sits on top of the agents you already use. Think of it as a universal remote for your AI coding workflow.

Here's the core idea: instead of managing Claude Code in one terminal, Codex in another, and Cursor in a third window, T3 Code gives you a single, polished interface across four platforms:

• iOS app — Full agent control from your iPhone
• Android app — Same capabilities on Android devices
• Web app — Browser-based at app.t3.codes, works from any machine
• Desktop app — Electron-based native app for macOS, Windows, and Linux

The server runs on your development machine (where your code, git repos, and provider CLIs live), and every client connects to it. Your agents keep running on the machine where the code is — T3 Code just gives you eyes and hands from anywhere.

Supported AI Coding Agents

T3 Code supports five major AI coding agent providers, each with their own CLI that T3 Code wraps:

Provider	CLI Required	Authentication
OpenAI Codex	Codex CLI	codex login
Claude Code	Claude Code CLI	claude auth login
Cursor	Cursor CLI	agent login
Grok Build	Grok Build CLI	grok login
OpenCode	OpenCode CLI	opencode auth login

Getting started takes about 60 seconds:

# Install and launch T3 Code (requires Node.js 22.16+)
npx t3@latest

# That's it. The server starts and opens the local web app.

For a more permanent setup, native packages are available:

# macOS
brew install --cask t3-code

# Windows
winget install T3Tools.T3Code

# Arch Linux
yay -S t3code-bin

Remote Access: Control Your Agents From Anywhere

This is where T3 Code earns its stars. The remote access architecture supports three connection methods, from simple LAN pairing to production-grade Tailscale mesh networking:

1. LAN Pairing (Simplest)

Both devices on the same network? Just pair them:

# On your development machine:
npx t3 pair

# A QR code appears — scan it with your phone's T3 Code app.
# Done. You're connected over your local network.

2. Tailscale (Recommended for Remote)

For working from anywhere without exposing ports to the internet:

# Pair over your tailnet — no port forwarding needed
npx t3 pair --tailscale

# The server publishes over Tailscale Serve HTTPS
# Your phone connects via https://your-machine.tailnet.ts.net/

Tailscale integration gives you transport-layer security, stable MagicDNS addresses, and zero public internet exposure. The desktop app auto-detects Tailscale and offers tailnet endpoints in the connection settings.

3. SSH Launch (Advanced)

For developers who want the desktop app to manage remote servers:

# In the desktop app:
# Settings → Connections → Remote Environments → Add environment
# Enter SSH target: user@your-server.com
# T3 Code probes the host, starts a remote server, opens a local port forward

The SSH launcher writes a small script under ~/.t3/ssh-launch/, starts or reuses a remote T3 server, and forwards the port back to your desktop. Your remote machine still owns the files, git state, and provider sessions.

Permission Modes: Balancing Speed and Safety

T3 Code implements four permission modes that control how much autonomy your AI agents have. Each mode is set per thread, so you can run one task fully autonomous while supervising another side-by-side:

• Supervised — Ask before every command and file change. The agent pauses, shows what it wants to do, and waits for your approval. Best for unfamiliar tasks or production repos.
• Auto-accept edits — File changes go through automatically; shell commands still require approval. Great for refactoring sessions where you trust the code changes but want to watch the build commands.
• Auto — Routine actions proceed without prompts; risky ones still ask. Implementation varies by provider: Codex uses an AI reviewer, Claude uses its own auto-mode, and others fall back to asking.
• Full access — The agent runs unattended until it finishes or asks a question. Ideal for sandboxed worktrees and experiments.

Approvals appear inline in the conversation thread. You approve or reject with a tap, and the agent continues from where it paused. The mobile app offers the same four modes, so your phone has the same control granularity as your desktop.

Real-World Example: The Commuting Developer

Here's how T3 Code changes a typical workday:

8:00 AM — Desk: You start Claude Code on a complex feature branch. Set the permission mode to "Full access" and let it work through the implementation. Head to the kitchen for coffee.

8:15 AM — Kitchen: Check your phone. Claude Code has made progress but hit a decision point — it's asking whether to refactor the database schema or add a migration. You review the diff on your phone, approve the migration approach, and Claude Code continues.

8:45 AM — Commute: On the train, you switch to a new thread and ask Codex to write integration tests for the changes Claude Code made. Set to "Auto-accept edits" — you trust the test generation but want to see any build commands it runs.

9:30 AM — Office: Back at your desk, both tasks are complete. The test suite passes. You push the branch and open a PR — all from the same T3 Code interface.

Without T3 Code, steps 2 and 3 are impossible. You'd be tethered to your workstation or forced to set up complex SSH forwarding and terminal multiplexer configurations.

Key Benefits

• True multi-device control — iOS, Android, web, and desktop all connect to the same server with feature parity
• Provider-agnostic — Switch between Codex, Claude, Cursor, Grok, and OpenCode without changing your workflow
• Remote-first architecture — LAN, Tailscale, and SSH connectivity out of the box
• Granular permission control — Four modes per thread give you the right balance of speed and safety
• Open source — Fully open, MIT licensed, and forkable. If the team ever goes the wrong direction, you have everything to build your own version
• Zero-config setup — npx t3@latest gets you running in seconds, no complex configuration required
• Multi-account support — Run multiple Codex or Claude accounts simultaneously for different projects or clients
• Active development — Built by the ping.gg team with a growing community of 17,700+ stars and 4,000+ forks

Getting Started in 3 Steps

# Step 1: Install at least one provider CLI and authenticate
# Example with Claude Code:
claude auth login

# Step 2: Launch T3 Code
npx t3@latest

# Step 3: Start a session and pick your provider
# The web app opens automatically — create a thread and start coding

For remote access from your phone:

# Pair your mobile device
npx t3 pair

# Scan the QR code with the T3 Code mobile app
# You're connected — manage your agents from anywhere

Frequently Asked Questions

Is T3 Code free to use?

Yes, T3 Code is completely free and open source under the MIT license. You use your existing subscriptions to AI coding agents (Claude, Codex, Cursor, etc.) — T3 Code is just the control interface, not an AI service itself.

What Node.js version do I need?

T3 Code requires Node.js 22.16+, 23.11+, or 24.10+. The server runs on your development machine and needs a compatible runtime. Remote hosts used via SSH launch also need a compatible Node.js version.

Does T3 Code replace my existing IDE or editor?

No. T3 Code is a control surface for AI coding agents, not a code editor. You still use VS Code, Vim, or whatever editor you prefer for manual code changes. T3 Code specifically manages the AI agent workflow — sending tasks, reviewing output, and approving actions.

How secure is remote access?

T3 Code recommends using Tailscale for remote connections, which provides end-to-end encryption and zero public internet exposure. LAN pairing works over your private network. SSH launch uses your existing SSH keys and tunnels. T3 Code never asks you to expose ports to the public internet.

Can I run multiple AI agents simultaneously?

Yes. T3 Code supports multi-threaded conversations, so you can run Claude Code on one task while Codex works on another. Each thread has its own permission mode, and you can switch between them from any connected device.

What happens if I lose connection to the server?

Your AI agents keep running on the server machine. When you reconnect from any device, you pick up exactly where you left off. The server maintains all state — conversations, file changes, git status, and terminal sessions persist across disconnections.

Is T3 Code affiliated with Anthropic, OpenAI, or any AI provider?

No. T3 Code is built independently by ping.gg. It wraps the official CLIs from each provider but has no formal affiliation with any of them. It works with whatever subscriptions you already have.

Final Thoughts

The AI coding agent ecosystem is fragmenting fast. Every major AI company now has its own CLI agent, each with its own interface, permissions model, and workflow. T3 Code's bet is that developers want a unified control plane — not five different terminal windows and no way to check on progress from their phone.

With 17,755 GitHub stars and backing from one of the most respected developer tooling teams, T3 Code is positioning itself as the standard interface layer for the AI-augmented development workflow. Whether you're a solo developer managing multiple agents or a team lead wanting remote visibility into AI-driven code changes, it's worth adding to your toolkit.