T3 Code: One Control Plane for Claude Code, Codex, and Every Agent You Run T3 Code, built by Theo Browne's team at T3 Tools, crossed 59,000 GitHub stars in under 90 days as an MIT-licensed control plane that wraps existing coding agents — Claude Code, OpenAI Codex, Cursor, Grok Build, OpenCode, Antigravity, and any agent speaking the Agent Client Protocol — in one interface across desktop, web, iOS, and Android. The tool does not resell tokens or require a cloud account, and it adds per-thread git worktrees with one-button pull requests, mobile approval of running agents (thread opening 40% faster after optimization, median latency down from 577ms to 346ms), and Orchestrator V2 multi-agent delegation via delegate_task, stable since v0.1.0. T3 Connect remote access is free for up to three machines, and enterprise managed policies landed November 2, letting IT teams set and lock configuration across developer fleets. T3 Code just crossed 59,000 GitHub stars in under 90 days. That kind of growth doesn’t happen by accident — it happens when a tool solves a problem developers are actively fighting. The problem: you’re juggling three AI coding agents across two machines, approving steps from a terminal at 11 PM, with no way to manage diffs, switch models mid-task, or check progress from your phone. T3 Code, built by Theo Browne’s team at T3 Tools, is the unified control surface the agentic coding era has been missing. Not a New Agent. Not a New Subscription. This distinction matters. T3 Code is not a new AI model. It is not a new coding agent. It wraps the ones you already pay for — Claude Code, OpenAI Codex, Cursor, Grok Build, OpenCode, Antigravity, and anything that speaks the Agent Client Protocol ACP https://github.com/pingdotgg/t3code — and gives them a single, coherent interface across desktop, web, iOS, and Android. It does not resell tokens. It does not require a cloud account. It does not add a subscription layer. You bring your existing credentials; T3 Code is the control surface on top. The MIT license means you can fork it if the project ever takes a direction you don’t like. This is the right answer to subscription fatigue in the AI tooling space. Every new tool wants another monthly charge. T3 Code bets the opposite: be the best surface for agents developers already use. Five Features Worth Knowing 1. Unified Threads Across Every Agent All agents share one thread view. You can change models mid-thread and let Claude Code hand a subtask to Codex. Desktop, web, and mobile all show the same threads in real time. T3 Code treats dropped frames as bugs — performance is not an afterthought. 2. Mobile Approval Workflow Start a task at your desk. The agent keeps running on your machine. When it needs a decision, approve from your phone. T3 Code ships native iOS and Android apps — not web wrappers. After a recent optimization, opening threads on mobile is 40% faster median latency dropped from 577ms to 346ms . If you’ve ever wanted to push a production fix while away from your desk without opening a laptop, this is how. 3. Git Worktrees with One-Button PRs Each thread gets its own git worktree — a separate branch and working directory, sharing the same Git object database. T3 Code saves a checkpoint per agent turn, so the diff panel shows exactly what the latest turn changed, not everything uncommitted. If a turn goes wrong, revert the conversation and try again. When you’re satisfied, one button commits, pushes, and opens a pull request with a generated description. This works with GitHub, GitLab, Bitbucket, Azure DevOps, Forgejo, and Gitea. 4. Multi-Agent Delegation Orchestrator V2, stable since v0.1.0, lets a parent thread delegate subtasks to another model via delegate task . Claude Code can hand work to Codex mid-thread; T3 Code transfers conversation context when switching providers. Multi-agent workflows that previously required custom orchestration code now work through the same thread UI. 5. Remote Access and Enterprise Policies Three ways to reach your T3 Code host remotely: local network, Tailscale, or T3 Connect no router or VPN setup required . T3 Connect is free for up to three machines. Enterprise managed policies landed November 2: IT teams can now set and lock T3 Code configuration across an entire fleet of developer machines. Getting Started Try T3 Code without installing anything: npx t3@latest To install properly on macOS or Windows: macOS brew install --cask t3-code Windows winget install T3Tools.T3Code T3 Code needs at least one supported agent installed and authenticated. Install Claude Code, run claude auth login , and you’re ready. Full setup documentation lives on the T3 Code GitHub repository https://github.com/pingdotgg/t3code . How It Compares | | T3 Code | Merlin | OpenCode | |---|---|---|---| | Local-first | Yes | No hosted | Yes | | Mobile app | Yes | Slack | No | | Multi-agent | Yes | Limited | No | | Git worktrees | Yes | No | No | | Free | Yes MIT | Paid tiers | Yes | ByteIota covered Merlin https://byteiota.com/merlin-claude-code-codex-team-agent/ last week — it runs Claude Code and Codex as team agents from Slack and Linear. The two tools solve different problems: Merlin is a hosted team integration; T3 Code is a local control surface you own and run. If your workflow is Slack-first, Merlin may be a better fit. If you want full control over execution and a cross-device interface without giving up your existing subscriptions, T3 Code is the stronger answer. Caveats Worth Naming T3 Code is pre-1.0. v0.1.9 shipped November 11 https://github.com/pingdotgg/t3code/releases/tag/v0.1.9 and the project carries around 4,000 open issues. It moves fast — several nightly builds per day — which means stability can lag features. For remote access, the host machine must stay on and signed in; there is no cloud execution fallback. The agents run with their CLI’s full machine permissions: T3 Code adds reach, not a sandbox. The Bottom Line 59,000 stars in 90 days is not hype. Developers who run Claude Code or Codex are installing T3 Code because it solves a real workflow problem: too many terminals, too many diffs, no way to manage agent runs from anywhere but the original machine. The git worktree integration and mobile approval loop alone justify the install. If you run any supported coding agent, run npx t3@latest and see for yourself.