A third-party iOS app called Open Minis has shipped what amounts to a full agent platform on iPhone and iPad, using a sandboxed Linux shell and custom command-line interfaces to bridge frontier AI models to native Apple frameworks. MacStories editor Federico Viticci, who has reviewed iOS software for 17 years, called it “the most impressive indie app I’ve seen in a while.”

How It Works

Open Minis runs a version of iSH, an open-source Alpine Linux shell emulator, inside the iOS app sandbox. The developer built dedicated CLIs for each Apple framework: EventKit for Calendar and Reminders, MusicKit, HomeKit, HealthKit, Files, Vision (OCR), Apple NLP, clipboard access, speech recognition, NFC, and Bluetooth.

The result: a frontier model (GPT-5.6, Claude Fable, or any OpenRouter-compatible model) can issue tool calls through a Linux terminal to control native iOS functionality. The user sees a standard chat interface. The terminal runs in the background, optionally visible as a picture-in-picture preview.

The app supports MCP server execution, Python scripts, Git, SSH, and standard Unix utilities. It exposes its workspace through the iOS Files app as a native file provider, and can read from or write to any user-bookmarked folder across file providers.

Self-Modifying Agent Workspace

What separates Open Minis from typical AI chat wrappers is self-modification. Users can ask the agent, in natural language, to add new models, install MCP servers, create skills, or change app settings. The agent uses its own CLIs to inspect and modify its workspace configuration.

Viticci tested this by asking the app to install his open-source Apple Frames CLI from a GitHub link. The model found all dependencies, installed them, pulled the CLI’s skill definition, and was operational within a minute. He then asked it to grab three screenshots, randomize frame colors, merge them, and save the result to a specific folder.

Credentials are stored through native iOS Keychain integration rather than plaintext files.

The Siri AI Comparison

Viticci ran identical queries through Open Minis and Siri AI. When asked for a temperature and humidity report from HomeKit sensors, Open Minis parsed raw sensor dumps from every room and produced a formatted table. Siri AI’s response was, by Viticci’s presentation, significantly less useful.

A photo library query (“Based on my Photos, would you say I was recently on vacation?”) produced similarly divergent results. Open Minis cross-referenced its own conversation memory with photo metadata and HealthKit walking data to reconstruct a detailed trip narrative. Both apps use the same underlying Apple APIs.

Apple makes these frameworks available to all third-party developers, according to Viticci’s review. Open Minis achieves what Siri AI cannot by pairing that access with a frontier model’s reasoning and tool-calling capabilities.

Consumer Agents Reaching Platform Parity

Open Minis represents a specific trend: consumer-facing agent applications reaching feature parity with developer-oriented platforms. Viticci explicitly compared Open Minis to OpenClaw and Claude Code, noting that it achieves similar self-modifying, tool-calling agent behavior within the constraints of iOS app sandboxing.

The app ships today on the App Store. It requires users to bring their own API keys for model access.