Coding agents on your phone: Claude Code mobile, Codex, Cursor and open-weight models on Buildful

You can use a coding agent from your phone in two ways: drive one that runs on a computer you leave on, or hand the task to one that runs in the cloud. Claude Code mobile, Codex and Cursor each offer both. GitHub Copilot runs in the cloud, and open-source clients like Happy and OpenCode drive your own computer.

On Buildful you don't open an app at all: you text the task over iMessage, an open-weight model builds it in a cloud sandbox, and the pull request comes back by text. When I texted it a small README edit for our own repo, the pull request was open in under three minutes, and the repo's checks (npm run check: typecheck, 920 tests and lint) had run and passed.

A phone is good at handing off a task and reviewing the pull request. It's bad at running the code, the builds and the tests. So the question that sorts every option is where the code runs while your phone is in your pocket.

In order, this post answers:

• Which coding agents can I use from my phone, and does my laptop have to stay on?
• Can I run an open-weight coding agent from my phone, or only Claude and GPT?
• What is Buildful for iMessage, and how do I text it a task and get a pull request back?
• What can't it do yet?

I co-founded Hashnode and Bug0, which tests web apps with AI, and I'm building Buildful with Sandeep Panda. So weigh the product parts with that in mind.

The thread below is a real task I texted from my iPhone to our own repo, Hashnode/buildful.

I texted the task at 12:17 PM, and GitHub shows the pull request opened at 12:19. Buildful texts once when the task starts and again when the pull request is ready.

Every way to use a coding agent from your phone

There are two kinds: tethered agents, which need your computer on, and hand-offs, which don't.

A tethered agent runs on your own computer, and your phone only steers it. Your files and tools stay local, and the computer has to stay awake.

A hand-off runs the agent in a cloud sandbox, so you can close the laptop, and the agent only sees what's in your repo.

Claude Code, Codex and Cursor all have a real cloud option, so "the laptop has to stay on" isn't true of any of them as a whole. It's true of the two Remote Controls, of Codex on a connected computer, and of Claude Code's channels, which deliver your text to a session that's already open on your machine.

Anthropic's own mobile guide puts it plainly: if your computer will be off, use cloud sessions.

A tethered agent is the right pick when the task needs something only your computer has: a local database, a tool that isn't in the repo, or an MCP server you run yourself. For everything else, a hand-off is simpler: the agent keeps coding and testing while your phone sleeps.

Can you run an open-weight coding agent from your phone?

Yes, and it's the part most phone guides skip.

Open-weight models publish their weights, so anyone can run them. Most developers call them open source. Kimi K3, GLM 5.3 and DeepSeek V4.1 Flash are all open-weight.

There are two ways to get one on your phone. The open-source route is OpenCode or Happy on your own computer, connected to an open-weight model you pay for yourself, through OpenCode Go or a provider's API.

It works, and it's tethered: your computer stays on, and you set up the tunnel and the password yourself.

The other route is Buildful, where the open-weight model runs in the cloud. By default, Smart Mode picks the right open-weight model for each task, a fast one for a small fix and a stronger one with more time to think for a hard task.

You can also pick a model by name. Kimi K3, GLM 5.3 Flash, DeepSeek V4.1 Flash and MiniMax M3.1 Flash are all in the picker. The Buildful Pass, a $10 a month add-on to Pro or Team, gives you $80 of usage a month on those four and Particle-1.

Frontier models like Claude and GPT are there whenever you choose one, paid from prepaid credit (from $5) or your own API key.

A phone keyboard keeps tasks short, so what you send from a phone is small and frequent: fix this redirect, change that label, add the missing empty state. Because those tasks run on open-weight models, they cost about a cent each.

The open-weight models page has the full list, and the Pass post has every model you can pick.

Introducing Buildful for iMessage

You can text a task to Buildful from your iPhone and get the pull request back by text.

We announced it in the changelog, and this post is the longer version.

We built it because a text is the simplest way to hand off a task from a phone. There's no app to install, no new sign-in and no terminal session to keep open. You text Buildful the way you'd text a teammate.

Behind the text it's the same Buildful as in the app. Every task runs in a private sandbox that's deleted when the task ends, your checks (lint, typecheck, tests) run on every task, and new features and fixes that touch the UI get QA testing in a real browser, with a video recording of the test.

Smart Mode picks the open-weight model unless you've chosen another model for your phone, and a texted task counts against your plan's tasks like any other. iMessage joins Slack and Linear as ways to start a task without opening Buildful.

Step 1: Link your phone to Buildful

You link your phone once, from the Integrations page.

1. In Buildful, open Integrations and click Connect on the iMessage card.
2. Buildful shows a one-time code. Text it to the Buildful number from your iPhone.
3. Buildful replies that you're connected, with an example of a task to send.
4. Choose what runs the tasks you text: Smart Mode, or a model you pick. Texted tasks use the same plan and billing as tasks you start in the app.

Linking is personal. Tasks you text are filed under you, even on a team, and each teammate links their own phone.

If you haven't told Buildful how to check your code yet, add your repo's setup and check commands on the Repos page, as in step 1 of the Linear post. Buildful runs those checks on every pull request.

Step 2: Text the task

Write it like a good issue, in one text, and name the repo.

Say what you want, where, and what done looks like. "Fix the login redirect in acme/web" is the example Buildful sends when you link your phone, and it's the right shape. A text like "the login is broken" makes the agent guess.

Buildful replies "Starting your task in" and the repo name, with a link to the task. You don't need the exact repo name. In the thread at the top I wrote "in the buildful repo", and Buildful started the task in Hashnode/buildful.

In iMessage the link previews as "A private Buildful task", because only your team can open it.

If your text doesn't name a repo and Buildful can't tell which one you mean, it replies with a numbered list, and you answer with a number. A team with one repo is never asked.

One text starts one task, so send two texts for two tasks. They run at the same time, each in its own sandbox.

Step 3: Review and merge from your phone

The proof comes to your phone: the results of your checks and, for UI work, a video recording of the QA testing.

When the task is done, Buildful texts "Your PR is ready" with the link. The pull request has a summary of what the agent built, a self-review of the code and the results of your checks.

If the task touches the UI, Buildful runs QA testing in a real browser and makes a video recording of it. The pull request on GitHub shows a short GIF of that recording, and the full video opens in Buildful after you sign in. Both work on a phone, and so does merging in the GitHub app.

On my README task, the pull request said it was built with Smart Mode and that npm run check passed, 920 tests with no lint errors. It also said it skipped QA testing in the browser, because a README edit has no UI to test. That's the kind of note I want before I merge.

If the pull request needs a fix, open the task in Buildful on your phone and describe it. The new commits land on the same pull request, and Buildful texts you again when that turn is done.

I wrote more about what to check before you merge in the Linear post.

What it can't do yet

• Every text starts a new task, so you can't reply to a finished one by text. Follow-ups happen in the Buildful app.
• Buildful reads the words in your text and skips photos. A screenshot of a bug doesn't reach the agent, so describe the bug instead.
• iMessage needs an iPhone. On Android, start tasks from Slack with @Buildful, or from the Buildful web app in your phone's browser.
• You get a text when the task starts and one when each turn ends, but nothing about the steps in between. The task link shows those live.
• Group chats are ignored.

How Buildful compares with the other phone options

Claude Code's cloud sessions, Codex Cloud, Cursor's cloud agents and Copilot all keep working with your laptop closed, and most of them can run your tests.

On Buildful, you start the task with a text, with no app to install and no computer left on. It runs on the right open-weight model unless you pick another, at 1,000 tasks for $10 a month.

The pull request arrives with your checks run and, when the task touches the UI, a video recording of the QA testing, so you can decide from your phone whether to merge.

If you're weighing one of them, the Claude Code, Codex, Cursor and GitHub Copilot comparisons go into more detail, and all comparisons are in one place.

Frequently asked questions

Is there a Claude Code mobile app?

No, there's no separate app. Claude Code lives in the Code tab of the Claude app for iOS and Android. From there you can start cloud sessions, which run on Anthropic's infrastructure, or drive a session on your computer through Remote Control.

Can I use Claude Code from my phone with my laptop closed?

Yes, with cloud sessions, which need your repo on GitHub and a claude.ai plan. Remote Control and Dispatch drive your own computer, so it has to stay on with Claude Code or the Desktop app running.

How do I control Claude Code from my phone?

Run claude remote-control on your computer, or /remote-control inside a session that's already open. Then scan the QR code the terminal shows, or open the Code tab in the Claude app and pick the session. The session keeps running on your computer.

Can I use Claude Code on Android?

Yes. Claude Code is in the Code tab of the Claude app for Android, the same as on iPhone, with cloud sessions and Remote Control.

Does Codex work on mobile?

Yes. Codex mobile lives in the ChatGPT app for iOS and Android. You choose Cloud and an environment, or a Mac or PC you've connected, and describe the task. A connected computer has to stay awake and online.

Does Cursor work on mobile?

Yes, with the Cursor app for iPhone and iPad, or cursor.com/agents in any browser. Cloud agents keep running with your laptop closed and need Premium or a higher plan. Remote Control drives a session on your computer, which has to stay awake. An Android app is planned.

Is there an open-source coding agent I can use from my phone?

Yes. OpenCode is open source and runs a web interface with opencode web, which you can reach from your phone's browser over your network or a tunnel. Happy is an open-source mobile app for Claude Code and Codex. Both drive an agent on your own computer.

Can I use open-weight models from my phone?

Yes. With OpenCode on your own computer, you connect an open-weight model and pay its provider. On Buildful, Smart Mode picks an open-weight model for each task in the cloud, or you pick one by name, such as Kimi K3 or GLM 5.3 Flash.

Which open-weight model is best for coding?

It depends on the task, which is why Smart Mode picks one per task. If you'd rather choose, the Buildful Pass post lists every model you can pick, with which ones are open-weight.

Can I text a coding agent?

Yes, two ways. Claude Code's iMessage channel, a research preview, sends your text to a Claude Code session running on your Mac. Buildful's iMessage integration starts a task in the cloud and texts you the pull request, so your own computer can stay off.

What is Buildful for iMessage?

It's a way to start Buildful tasks by text. You link your iPhone on the Integrations page, text what you want built, and Buildful builds it in a cloud sandbox and texts you the pull request. It shipped on October 29, 2026, as the changelog entry describes.

Can I try Buildful before paying?

Yes. Your first 3 tasks are free, with no card, and they run on Smart Mode. Pro is $10 a month for 1,000 tasks, and Team is $40 a month for 3,000 tasks shared across the team. The plans are on the pricing page.

Start building on Buildful, then text it your first task.