OpenMinis/OpenMinis

OpenMinis – On‑device AI agent for iOS & Android

What it is – OpenMinis is an open‑source mobile application that lets you run large‑language‑model assistants (Claude, GPT, Gemini, etc.) directly on your phone. The app bundles a sandboxed Alpine Linux shell, so the agent can install packages, run scripts, and manipulate real files, while also exposing native device services (Health, Calendar, HomeKit, Bluetooth, clipboard, media, etc.) as tools the model can call.

Core capabilities

Feature	What it gives you
Bring‑your‑own model	Connect any LLM provider via API key or account sign‑in.
Linux sandbox	A sandboxed Alpine environment (iSH on iOS, PRoot on Android) runs on the device; the agent can install packages, run shell commands, and keep a persistent filesystem.
Deep device integration	Direct access to health data, calendar, contacts, HomeKit, Bluetooth, clipboard, media playback, alarms, etc.
Browser automation	The agent can open web pages, scrape content, and interact with sites on your behalf.
Skills & memory	Re‑usable “skill” folders (a SKILL.md plus optional scripts) are loaded on demand; the agent retains persistent memory across sessions.
Workspaces	Separate contexts addressed via minis://workspace/ URLs, letting you organise different projects.
Native offloads	Heavy or platform‑specific tasks can be delegated to compiled Swift/Kotlin code instead of the sandbox.

Typical user workflows

• Nutrition logging – Snap a meal photo, the agent recognises dishes, estimates calories/macros, and writes the data to Apple Health.
• Morning briefing – A Shortcut triggers Minis to fetch your X/Twitter timeline, summarise it, synthesize speech, and play it as an alarm.
• Task extraction – Pull messages from a Telegram group, have the agent turn bugs and action items into Apple Reminders, deduplicating automatically.
• Obsidian vault sync – Use a workspace to read, edit, and write Markdown notes directly into an Obsidian vault.
• Share‑to‑event – Send any page or message through the iOS Share Sheet; Minis creates a calendar event with time, location, and description.

Ecosystem

• AwesomeMinis – A community‑curated list of use‑cases and workflows.
• MinisSkills – A repository of ready‑made skill packages (TTS, web search, media download, health analysis, cloud APIs, etc.) that the agent can load on demand.

Technical stack

• iOS – Swift / SwiftUI app, with extensions for sharing, widgets, and file‑provider integration.
• Android – Kotlin / Jetpack Compose app, JNI bridge for native code.
• Sandbox – iSH (ARM64 fork) on iOS, PRoot on Android, both built from source; Alpine Linux rootfs provides the userland.
• Media handling – FFmpeg and LAME compiled into the sandbox for audio/video processing.
• LLM connectivity – Uses provider SDKs or raw HTTP calls (e.g., SwiftAnthropic for Claude, standard OpenAI API for GPT, etc.).

Building – The repo contains scripts to compile the native dependencies (FFmpeg, LAME, iSH, PRoot) and to prepare the Alpine rootfs. After cloning with submodules, the README outlines platform‑specific steps for iOS (Xcode project) and Android (Gradle). Detailed instructions live in BUILDING.md.

License – GPL‑v3 (the app links GPL‑licensed components, so the whole distribution is GPL‑v3). Third‑party licenses are listed in THIRD_PARTY_LICENSES.md.

Community & contribution

• No pull requests accepted in this private mirror; issues are the primary feedback channel.
• Public discussion on Telegram, plus separate repos (AwesomeMinis, MinisSkills) that do accept contributions.
• Beta testing via TestFlight (iOS) and GitHub APK releases (Android).

Bottom line – OpenMinis is a fully open‑source, on‑device AI assistant platform that combines large‑language‑model chat with a real Linux environment and deep mobile OS integration, enabling powerful, privacy‑preserving automations on iPhone and Android.