T3 Code

T3 Code is an open-source "agent harness control surface" built by the T3 tools team. It lets you control the coding agents on your machine, including Claude Code, Codex, Cursor, Grok Build, and OpenCode, from an Electron desktop app, a web app, or native iOS and Android apps. The project is MIT licensed, intentionally open, and designed so you can fork it if the product ever moves in a direction you dislike.

Key Features

• Multi-agent control: Manage sessions across Claude Code, Codex, Cursor, Grok Build, and OpenCode without switching terminals.
• Remote-ready apps: Native mobile apps, a web app, and a desktop app can connect to the local server running on your machine.
• Permission modes: Fine-grained controls over what each agent may do before an action is approved.
• Source control integration: Built-in support for source-control workflows and project context.
• Project icons and keybindings: Customize project-level presentation and keyboard shortcuts for fast switching.
• Simple server model: npx t3@latest launches the backend and local web app with no paid account.

Use Cases

• Agent-heavy developers: Developers who run multiple coding agents and want one surface to start, pause, and review sessions.
• Remote work: People who want to check or continue an agent session from a phone while away from the desk.
• Open-source advocates: Teams that want a transparent, forkable alternative to closed agent desktop apps.
• Subscription users: Users who already pay for Claude, ChatGPT, or another agent subscription and want to drive those agents from a consistent UI.

Pricing

T3 Code is free and open source under the MIT license. You still pay for the underlying agent subscriptions you connect, such as Claude Code, Codex, Cursor, Grok Build, or OpenCode.

Getting Started

1. Install and authenticate at least one supported agent CLI, for example codex login, claude auth login, or opencode auth login.
2. Run npx t3@latest from a terminal with Node.js 22.16+, 23.11+, or 24.10+.
3. Open the local web app, or install the desktop or mobile app.
4. Connect your agent session and start controlling it remotely.

Frequently Asked Questions

Does T3 Code sell my data?

No. The project states it was built for the best possible agent development experience, it is open source, and sessions run through your local server.

What is the main risk?

The project is early and explicitly says to expect bugs. It also routes consumer agent subscriptions through third-party harnesses, which may violate a provider's terms, so check your provider agreement before relying on it.

Which agents are supported?

Claude Code, Codex, Cursor, Grok Build, and OpenCode are supported at launch, with documentation for multiple accounts per provider.

Alternatives

• Claude Code: Anthropic's terminal coding agent with first-party harness capabilities.
• Codex CLI: OpenAI's open-source coding agent CLI.
• OpenCode: An open-source coding agent for multiple model providers.
• Conductor: Another agent orchestration surface for coding workflows.

Tips & Best Practices

1. Start with one provider: Authenticate one agent first, verify the local server, then add more.
2. Review permission modes: Configure approval levels before delegating long-running tasks.
3. Keep the server updated: Sync the desktop or mobile app with the local server version to avoid stale connections.

Conclusion

T3 Code gives agent-heavy developers a unified, remote-ready control surface for the most popular coding agents, and its MIT license keeps the project auditable and forkable. It is worth trying for anyone juggling multiple agent CLIs, as long as you treat it as an early-stage tool and keep provider terms in mind.