T3 Code: A Control Panel Across All Coding Agents — BYO Subscriptions, Open Source, Mobile-First
> Subject: pingdotgg/t3code
> Crawled: 2026-10-03 ｜ Stars that day: 251
> License: MIT ｜ Language: TypeScript
Not Another Coding Agent — a Remote Control for All of Them
T3 Code's README opens with its positioning: "agent harness control surface" — a control panel for agent harnesses.
It is not Claude Code, not Codex, not Cursor, not Grok Build, not OpenCode, not Google Antigravity. It is a unified control layer across all of these coding agents. Whichever agents you have installed on your machine, T3 Code can control them.
This is fundamentally different from existing agent tools:
- Claude Code is an agent you talk to that writes code
- Cursor is an IDE with built-in AI agents
- Conductor is a desktop GUI wrapper for Claude Code
- Claude Desktop is Anthropic's desktop app
- T3 Code is a control panel across all agents — whichever one you use, T3 Code can manage it
- Claude Code? Install the Claude Code CLI yourself, run
claude auth loginyourself - Codex? Install Codex CLI,
codex login - Cursor? Install Cursor CLI,
agent login - Grok Build? Install it,
grok login - OpenCode? Install it,
opencode auth login - Google Antigravity? Enable it in Settings
- iOS app
- Android app
- Web app
- Electron desktop app
- Codex: OpenAI's coding agent
- Claude Code: Anthropic's coding agent
- Cursor: independent AI IDE
- Grok Build: xAI's coding agent
- OpenCode: open-source coding agent
- Google Antigravity: Google's coding agent (announced at Google I/O 2026)
- "We are very very early. Expect bugs."
- "We're (basically) not accepting contributions yet. Small fixes considered; large features no."
- Start an agent task on your computer
- T3 Code's background service monitors it
- Push notifications on your phone
- Review results and give feedback from your phone
- Not an agent (provides no models)
- Not an IDE (no code editing)
- Not a SaaS (no subscriptions sold)
- A control surface (cross-agent control layer)
Think of the relationship between a TV remote and the TV: T3 Code doesn't produce content (it provides no models); it just makes it easy to switch channels, adjust volume, and set timers.
BYO Subscriptions: Selling Nothing, Only the Experience
T3 Code's business model is counterintuitive: it sells nothing. The README states plainly: "Wait, what are you selling me? Nothing."
Users bring their own subscriptions (BYO):
T3 Code doesn't pay your API bills, manage your subscriptions, or run models. It does one thing: lets you control all of these agents through a unified interface.
The positioning is smart. The coding agent market is highly fragmented — Claude Code, Codex, Cursor, and Grok Build each have their own CLIs, interfaces, and shortcuts. Developers often use 2–3 agents simultaneously (e.g., Claude Code for writing, Codex for review), and switching between them is painful.
T3 Code doesn't try to replace any agent; it aims to be a unified entry point for all agents. This avoids direct competition with Anthropic, OpenAI, and Google while solving a real developer pain point.
Mobile-First: An Underrated Differentiator
T3 Code's most interesting differentiator is being mobile-first. It offers:
Most coding agent tools are desktop-only — because "writing code" happens on a computer. What's the logic behind mobile apps?
The answer: coding agents aren't just about "writing code". A full agent workflow includes: 1. Starting a task ("fix this bug for me") 2. Monitoring progress (the agent is running; check where it is) 3. Reviewing results (the agent finished; check the code) 4. Giving feedback ("wrong direction, try another")
Steps 2, 3, and 4 don't require being at your desk. You can check agent progress on your commute, review code between meetings, and give feedback at lunch.
This mirrors the mobile app logic of CI/CD tools — Jenkins and GitHub Actions both have mobile notifications, because "monitoring" and "reviewing" are mobile scenarios while "writing" is a desktop scenario. T3 Code extends that logic to coding agents.
"If we ever go the wrong direction, we want you to have everything you need to fork"
One README line stands out: if the project ever goes the wrong direction, the team wants you to have everything you need to fork and build the editor you want.
That's a strong commitment. Most open source projects say "contributions welcome"; T3 Code says "if you don't like our direction, fork it and build your own." The reasoning:
1. T3 Code doesn't try to lock users in. Open code, open formats, fork-friendly. 2. The team isn't 100% sure of its own direction. "We might be going the wrong way" is an honest admission. 3. "Users can leave" is a design goal, the exact opposite of the "users can't leave" SaaS logic.
This attitude matches Theo (pingdotgg)'s developer philosophy. Theo created the T3 Stack (Next.js + tRPC + Tailwind + Prisma) and has long championed "developer experience first." T3 Code continues that: tools should serve users, not lock them in.
"Currently supports Codex, Claude, Cursor, Grok Build, OpenCode, and Antigravity"
T3 Code's supported-agent list is itself a market map:
These six represent the market's major players, unified in one control panel — launch, monitor, and switch between any of them. This hints at a trend: the coding agent market won't have a single winner. Developers will use different agents for different tasks — Claude Code for refactoring, Codex for review, Cursor for completions. A cross-agent control layer becomes more valuable over time.
"We are very very early in this project. Expect bugs."
The README's honesty is notable. T3 Code explicitly says:
That's the opposite of the usual "contributions welcome" stance. The team chose to nail down the direction first, then open up contributions. The logic:
1. Early external contributions dilute direction. Every contributor has their own ideas; the project becomes a hodgepodge. 2. The team wants to validate its architecture first. If v1's architecture is wrong, retrofitting later is extremely expensive. 3. "Small fixes yes, big features no" is a pragmatic compromise — not rejecting community involvement, but not letting the community set direction.
This resembles the "full rewrite" logic of Cloudflare OS v2 — early projects would rather move slowly and get the architecture right.
An Underrated Detail: Background Service
T3 Code ships a t3 service install command that runs T3 Code as a background service. It's not a "launch when needed" tool but a continuously running agent monitoring service.
Same logic as Slack or Discord running in the background — you don't launch Slack every time you want to send a message; it's always running and notifies you. T3 Code extends this to coding agents: the agent runs, T3 Code monitors in the background, and notifies you of results.
Combined with the mobile apps, this forms a complete "coding agent monitoring" workflow:
This is nearly identical to CI/CD notification workflows — just with "builds/tests" swapped for "agent tasks".
Market Positioning
T3 Code's positioning is unique:
T3 Code wants to be the orchestration layer for coding agents. If the market stays fragmented — Claude Code, Codex, Cursor, Grok Build each holding a niche — a cross-agent control layer becomes a necessity.
---
Repository: https://github.com/pingdotgg/t3code
Install: curl -fsSL https://t3.codes/install.sh | sh
> Daily GitHub Trending Deep Dive · Post #85