How to Let ChatGPT Control Your Android Phone via Shelly
Learn how to pair ChatGPT with Shelly, an open-source Android terminal, to let AI control your phone while requiring manual approval for every single action.
Stock photo for illustration only, not from the actual event
- Shelly is an open-source Android terminal allowing ChatGPT to control device tasks.
- Every phone action strictly requires manual approval from the user.
- Setup uses device pairing without needing an API key, PC, or root access.
- Shelly includes a full terminal environment with bash, Python 3, and Node.js.
A developer has shared a method enabling the ChatGPT app to operate Android smartphones using its Codex mode paired with an open-source terminal application called Shelly. This setup allows ChatGPT to run terminal commands and manage app tasks directly on the device through a secure pairing process using the user's existing ChatGPT subscription.
Everything required for this operation runs directly on the smartphone itself. Users do not need a personal computer, Termux, or root privileges to get the system up and running.
Integrating AI with mobile hardware often raises security concerns, but this project addresses them by physically separating the authorization step from the AI model, ensuring that external web pages cannot inject hidden malicious commands.
Stock photo for illustration only, not from the actual event
The initial setup process involves the following steps:
- Install the APK and grant "All files access" permission upon first launch.
- Open the Shelly terminal and type codex to sign in via the built-in browser pane.
- Run shelly codex pair to generate a device pairing code for the ChatGPT app.
- Open the ChatGPT app, navigate to Codex, add a remote machine, and enter the pairing code.
Once configured, users can prompt ChatGPT to check available storage or execute tasks inside other applications, such as opening Spotify and saving a playlist. In the background, ClosePaw operates on a virtual display while the main screen remains accessible to the user.
"Prompt rules like never approve yourself aren't a security boundary. The approval has to live somewhere the model can't reach."
Ryo Itabashi
Security is enforced by requiring the user to manually unlock the device and tap approve on a notification for every single action. Prompt-based rules alone are insufficient for security, meaning the approval mechanism must reside completely outside the reach of the AI model.
Beyond remote control, Shelly functions as a full terminal environment supporting bash, git, Python 3, Node.js, ssh, tmux, and vim, along with local LLMs and scheduled agents. Source code and releases are publicly available on GitHub for developers and enthusiasts.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment