For a quick introduction to the web interface, you can launch it via npm if Node.js is installed:

npx t3@latest

On an Intel Mac, it is better to get the desktop build: there is no ready-made CLI for this architecture yet.

What to keep in mind for now

Models, tools, and some behavior still depend on the provider. When switching agents, the conversation context is passed along, while the internal states of different agents remain separate.

In the FAQ, the team reports an issue with the Anthropic Agent SDK: Claude’s five-hour limit is used up faster when working through the SDK than directly through Claude Code. According to the same report, the issue does not affect the weekly limit. If you plan to work with Claude through T3 Code all day, this may affect your plans.

Four parallel assistants may use up the available request allowance faster. It makes sense to give them the same task when you need several independent results.

So, is it a killer app after all?

If you work with a single agent in a single project, you may not need to switch. Your familiar app already solves your task, and you would have to learn a new interface.

T3 Code is worth trying if you already have to switch between subscriptions, manage parallel tasks, or monitor work across multiple machines. Here, you can bring agents together, assign them work, and review the results in a single queue. In such scenarios, it can compete with Claude Desktop and Codex Desktop.