T3 Code on iOS and Android: what the mobile app does

What ships, and where

T3 Code’s mobile clients arrived alongside T3 Connect, and the announcement framed them together: run npx t3 connect in a terminal, install the app, and you can drive Claude and Codex remotely. Both are free, both are published by T3 Tools Inc, and neither carries an in-app purchase.

What you can actually do from the phone

The scope is broader than most companion apps in this category, which usually stop at reading a transcript. T3 Code’s mobile client aims to be the desktop surface at a smaller size.

• Start agent runs. Brief an agent and kick off a thread against a repository on the connected machine.
• Browse files. Navigate the tree on the host rather than only reading what the agent quoted.
• Review diffs. Read the change the agent produced, which is the reason to open the app at all.
• Version control actions. Commit, push, and the source-control operations the desktop thread surface offers.
• Use a terminal. A real terminal on the host, which is the difference between diagnosing a stuck run and waiting to get home.

The terminal is the item most worth calling out, because it is the difference between a companion app and a real client. Agent runs fail in ordinary ways: a dependency is missing, a port is occupied, a test needs a service that is not up. Without a shell on the host, every one of those turns into a note to fix it later. With one, most of them are a thirty-second detour on a train.

The honest caveat is the one that applies to the whole project: this is alpha software on a nightly cadence, and mobile is where a regression is least convenient because you find it when you are away from the machine that could fix it. Treat the mobile client as excellent for supervision and steering, and keep the expectation that a genuinely broken session may need the desktop.

What a phone is actually good for

It is worth being precise about the job, because "code from your phone" oversells it and then disappoints people. Nobody writes a feature on a 6-inch screen. What a phone is genuinely good at is the supervision loop around a run that is already going, and that loop has four moments.

1. Kick off. You thought of the task on the way somewhere. Briefing an agent takes one paragraph, and the twenty minutes it then works are twenty minutes you were not going to use anyway.
2. Unblock. The agent asks a question or hits a permission it needs. This is the moment that decides whether remote control was worth setting up: an answer in thirty seconds versus a run that sat idle until you got home.
3. Read the diff. Not a full review, but enough to know whether the approach is right before more work is built on it. A phone is fine for reading a two-file change and useless for a thirty-file one.
4. Land it. Commit, push, open the pull request, and let the real review happen on a real screen with real reviewers.

The second moment is the one to test before you commit a workflow to any mobile client, and it is the one feature lists cannot describe. What matters is not whether the app can show you a question; it is whether you find out there is a question without opening the app. That is a notification and platform-integration problem rather than a client-features problem, and it is where the two products in this comparison diverge most sharply.

Connecting a phone

There are two routes, and which one you want depends on whether you are willing to have traffic cross a hosted relay.

T3 Code mobile versus Continuum on iPhone

On Android there is no comparison to make: T3 Code has a client and Continuum does not. If your phone runs Android, that decides it, and the rest of this section is not for you.

On iPhone the comparison is real, and it comes down to cross-platform reach against native depth. T3 Code’s app is built to be the same product on both mobile platforms, which is exactly why it exists and also what it costs. Continuum ships a native iPhone app with an Apple Watch companion and takes the platform features that follow from that choice.

The rows that change a workflow are the notification ones. A mobile client that requires you to open an app, find a thread, and read a screen is useful when you remember to check. A client that puts run progress on the Lock Screen and lets you clear a plan approval from a notification is useful without you remembering anything, which is the difference between supervising a twenty-minute run and discovering it stalled nineteen minutes ago waiting for a yes.

Which one to install

Nothing stops you from installing both. They are free, they drive the same agent CLIs, and they connect to machines rather than to each other, so a week of using each against real work will settle it faster than any table. If you are still deciding on the desktop side, what T3 Code is covers the architecture and the migration guide covers what moving actually involves, including the reasons to stay.

One last note on how to judge a mobile client, because it is easy to evaluate the wrong thing. The screenshots all look similar, and every app in this category can render a diff. What separates them is behaviour you only see over a week: whether the connection survives a subway tunnel, whether a run that finished while your phone was locked tells you so, whether opening the app after four hours shows you the current state or a stale one, and whether a permission prompt from the agent reaches you as something you can answer rather than something you have to go looking for. Install whichever one you like the look of, then judge it on a Tuesday afternoon rather than on a feature table. The same test applies to running Claude Code from a phone in general, whatever client you put in front of it.

Questions people ask

Keep reading

Continue through Claude Code reference with Switch from T3 Code to Continuum without losing work.