OpenMinis embeds a full Linux shell for AI agents on your phone

OpenMinis is a free, open-source AI agent app that runs directly on your iPhone or Android device. Instead of being just another chatbot wrapper, it bundles a full Alpine Linux sandbox right inside your phone and hooks directly into your mobile operating system.

You simply bring your own API keys or accounts from Claude, GPT, or Gemini. The model gets an actual Linux shell capable of installing packages, executing Python scripts, handling local files, and automating tasks across your phone.

What you can do with it

By marrying language models to on-device tools, OpenMinis turns messy manual workflows into automated background jobs.

You can snap a photo of your lunch and have the agent identify the dishes, estimate macros, and log them straight into Apple Health. You can set an iOS Shortcut so OpenMinis fetches your X timeline, summarizes the key posts, synthesizes audio, and plays it as your morning wake-up alarm.

It can also monitor Telegram groups to pull out bug reports, remove duplicates, and drop actionable items directly into Apple Reminders. If you use Obsidian, it mounts your local vault as a standard workspace to research, clean up, and save Markdown notes on device.

The app also supports extensible community workflows. It adopts the same SKILL.md progressive disclosure format popularized by Claude Code and Hermes Agent, meaning existing skill files written for desktop developer agents can run inside OpenMinis out of the box.

How it works under the hood

Getting a real Linux shell to operate reliably inside strict mobile operating systems required serious engineering trade-offs.

On iOS, OpenMinis runs a custom ARM64 fork of iSH with a threaded-code JIT interpreter, handling process management and over 100 system calls within the app's memory space. On Android, it takes a PRoot approach to manage a user-space chroot running Alpine rootfs without requiring root access.

Instead of bloating the model's context window with dozens of separate tool schemas, OpenMinis presents the LLM with only eight core tools. System integrations like calendar, contacts, HomeKit, and Bluetooth are implemented as native CLI utilities inside the sandbox. When the agent runs a command, the internal kernel intercepts the call and hands it off to native host code via JSON pipelines. This allows the model to chain mobile device operations using standard shell pipes and scripts.

To deal with iOS background execution limits, the app leverages location heartbeats and background audio along with Live Activities to display real-time progress.

How to try it

OpenMinis is completely open source under the GPLv3 license on GitHub. Android users can grab the prebuilt APK directly from the project's Releases page. On iOS, you can hop onto their TestFlight builds or compile the project locally using Xcode.

Why it matters

Most mobile AI attempts have stalled out as glorified chat screens or walled-off vendor features. OpenMinis flips the script by treating your smartphone like an actual computer and handing the terminal keys to an LLM. It shows that code execution and command-line interfaces are far more capable than static tool-calling frameworks on mobile.

Power users and developers who already have API credits should definitely give it a spin. Keep in mind that heavy background execution and sandbox emulation will put a dent in battery life, but as an independent showcase of what mobile agents can actually do, it is leagues ahead of standard chat apps.