VibePager · Guide
Vibe coding from your phone
Three quite different kinds of tool get called this, and picking the wrong kind is why people conclude it does not work. The question that separates them: where does the code live?
Updated September 2026
The term
Vibe coding, as Andrej Karpathy used it in early 2025, means leaning on an AI agent for the code and staying at the level of intent — describing what you want, reading the result, steering. The interesting part for phones is that if you are not typing the code yourself, the keyboard stops being the bottleneck. What you need instead is a way to read, decide and approve.
Three kinds of tool, one deciding question
Ask where the code lives. Everything else follows from the answer.
| Kind | Where the code lives | Good for | The catch |
|---|---|---|---|
| Phone-first builders (prompt-to-app services) | On their servers | Starting something from nothing, on a bus, with no setup at all | It is their environment, their stack and their deploy. Bringing an existing repository in is usually the part that does not work. |
| Browser IDEs (VS Code tunnels, code-server, cloud dev environments) | On your machine or a container you rent | Real editing, real terminal, everything your desktop has | It is a desktop IDE on a 6-inch screen. Fine for a quick fix, punishing for an hour. |
| Remote controls for your own agent | On your own machine, untouched | Continuing the actual project you were working on, with the agent and models you already pay for | Your computer has to be awake. There is no cloud copy working without you. |
Why the third kind exists
Because most vibe coding does not start from nothing. It starts from a repository you already have, an agent you already configured, and a task you were already halfway through when you had to leave the desk. None of that transfers to a prompt-to-app service, and putting a desktop IDE on a phone screen solves the wrong half of the problem.
What you actually do from a phone is smaller and more specific: send a task, read what the agent found, and answer it when it stops for permission. That is a handful of screens, not an IDE.
What good looks like on a phone
- It pages you. You should not have to check whether the agent is waiting.
- Decisions are one tap, with the diff visible. Approving blind is how people end up disabling permissions entirely.
- The conversation is continuous. Start at the desk, continue on the phone, finish at the desk.
- Your code stays yours. If a tool needs to upload the repository to be convenient, that is a trade you should make deliberately.
Where VibePager sits
Firmly in the third category. The agents keep running on your Mac or PC; the phone is a purpose-built control surface with push notifications, approval cards and a file browser. It does not host anything and it does not run models — it carries your decisions to the machine that is already doing the work.
Questions
- What is vibe coding?
- Working with an AI agent at the level of intent rather than syntax: you describe what you want, the agent writes and runs the code, and you review and steer. The term comes from Andrej Karpathy in early 2025.
- Can you actually vibe code on a phone?
- Yes, for the parts that matter: giving the task, reading the result and approving what the agent wants to do. What does not work well on a phone is typing code by hand — which is exactly the part you were not doing anyway.
- Do I need to be a developer?
- For a remote control, yes — it assumes you already have an agent set up on a machine. Prompt-to-app services are the ones aimed at people starting with nothing.
- Is there a free way to do this?
- SSH with tmux from a phone terminal app costs nothing and works, if you can live with the keyboard. Several prompt-to-app services have free tiers with limits.
VibePager runs the agents on your own machine and puts the controls on your phone.