PocketAI: Remote Control for AI Agents
May 2026
An Android remote control for your AI coding agents. Not for typing code on your phone, but for monitoring and steering long-running agents while they work on your desktop. Works with Claude Code, Gemini CLI, Codex, and open models.
What is it?
PocketAI is a remote control for your AI coding agents. The key word is remote control, not editor. It is not meant for typing code on your phone. It is meant for monitoring and steering a long-running agent while it does the heavy lifting on your desktop.
The scenario I kept running into was simple: I would kick off an agent on a real task, then have to leave my desk. Out of town, at the grocery store, on a walk, I had no idea what the agent was doing or whether it had gone off the rails. PocketAI fixes that. Check the progress from your hotel, see what your agent just did, and if it is heading the wrong way, send it a message and redirect it, all from your phone.
Which agents it controls
PocketAI is built to sit on top of the coding agents people actually use, not a single locked-in tool. It currently works with Claude Code, Gemini CLI, Codex, and free/open models, with Hermes already supported and more on the way.
The goal is that whatever agent you already run on your desktop, you can attach PocketAI to it and watch and steer it from your pocket. The phone is just a thin control surface. The real agent keeps running where it always was.
Privacy-first by design
The thing I cared about most was that this could not become another big-tech middleman sitting between you and your code. Your code stays 100% local. It never gets uploaded to my servers.
The only thing my relay does is establish the connection between your phone and your desktop. Nothing is stored and nothing is logged. The relay is a pipe, not a vault. For people who want to remove even that hop, Tailscale support is coming soon, so the phone and desktop can talk over your own private network with no relay in the middle at all.
Where to get it
PocketAI ships as an Android app on the Google Play Store. The Play Store link is the live link for this project.
If you run long agent tasks and hate being chained to your desk while they finish, that is exactly the itch this scratches.
Key takeaways
- Designing a mobile control surface for desktop agents instead of a mobile code editor
- Building a relay that only brokers connections — nothing stored, nothing logged
- Supporting multiple agents behind one interface: Claude Code, Gemini CLI, Codex, open models
- Privacy-first architecture: keeping code 100% local with optional Tailscale peer-to-peer
- Shipping and distributing an Android app through the Google Play Store