Open Minis

Your Private On-Device AI Agent

An on-device AI agent that goes beyond chat. It takes action — browsing the web, accessing your health data, managing your schedule, controlling your smart home, and automating complex tasks. All privately on your device.

Download on the App Store

Features

Multi-Model AI

Connect to Anthropic Claude, OpenAI GPT, Google Gemini, OpenRouter, or any OpenAI-compatible endpoint. Switch models per conversation.

Built-in Linux Shell

A full Alpine Linux environment runs on-device. The agent writes and executes scripts, installs packages, and processes data — no server needed.

Deep iOS Integration

Native access to HealthKit, Calendar, Reminders, HomeKit, Contacts, Bluetooth, Location, Photos, Speech, and more.

Web Browser

The agent browses the web, fills forms, extracts content, and takes screenshots — all within the app.

Custom Skills

Import or create AI skills to extend the agent's capabilities. Compatible with SKILL.md format.

Privacy by Design

API keys stay in your device Keychain. No data collected. No third-party analytics. Your conversations are yours.

Getting Started

1

Download & Install

Get Open Minis from the App Store. Requires iOS 16+, iPadOS 16+, macOS 13+ (Apple Silicon), or visionOS 1.0+.

2

Add an AI Provider

On first launch, you'll be guided to add your AI provider. Choose from the supported providers below and enter your API key.

3

Select a Model

Pick a model from your provider. You can add multiple providers and switch models per conversation.

4

Start a Conversation

Create a new session and start chatting. The agent can execute commands, browse the web, read your calendar, and much more.

Supported AI Providers

Open Minis supports multiple AI providers. You need at least one API key to get started.

Provider Models Get API Key
Anthropic Claude Opus 4.6, Sonnet 4.6, Haiku 4.5 console.anthropic.com
OpenAI GPT-4o, o-series platform.openai.com
Google Gemini Gemini 2.5 Pro/Flash, 3 Pro/Flash aistudio.google.com
OpenRouter Multi-provider routing openrouter.ai
Custom Any OpenAI-compatible API Your provider's dashboard

Advanced Configuration

Model Groups

Group multiple models together with routing strategies:

Agent Loop Models

Configure a separate pool of models that the AI agent can use when it needs to delegate sub-tasks or reason through complex operations.

Skills

Extend the agent with custom skills. Import a SKILL.md file from a URL or your file system, or let the agent create one during a conversation. Skills can be enabled or disabled per session.

Session File System

Each conversation session has an isolated file system mounted at /var/minis/ inside the Linux environment. Files are organized into namespaces:

Native iOS Capabilities

The agent has access to iOS frameworks through native offloads. Just ask naturally:

FAQ

API Configuration & Network

Why does my API key show "Invalid" or fail to authenticate?
  • Google Gemini OAuth has strict third-party restrictions and may block access. We strongly recommend using an API Key instead of OAuth for Gemini.
  • Check your balance — make sure your API account has credits or is within the free tier quota.
  • Network stability — Agent mode requires a stable connection. If you use a proxy or VPN, assign a fixed node for Open Minis traffic to avoid IP changes that trigger rate limits.
How do I add OpenAI-compatible providers (e.g. Qwen, MiniMax, DeepSeek)?

Most third-party models support OpenAI-compatible endpoints. In Provider settings, choose Custom (OpenAI-compatible), enter the base URL (without the trailing /v1/... path) and your API key. If the model list doesn't auto-populate, manually type the model ID (e.g. qwen-turbo, deepseek-chat).

How do I configure a third-party API proxy / relay?
  • Check your relay's documentation to determine if it's OpenAI-compatible or Anthropic-compatible.
  • In Provider settings, select the matching type, enter the base URL only (strip any trailing path like /v1/chat/completions) and your API key.
  • If models don't appear automatically, enter the model ID manually.

Model Selection & Usage

Which AI model works best for agent tasks?
Use CaseRecommended
General agent tasksClaude Sonnet / Opus (most stable tool calling)
Long conversation stabilityGPT-5.4 / Codex
Budget-friendly / backupKimi-k2.5, MiniMax, Qwen3.5
What is "Enhanced Cache" and when should I enable it?

This feature currently works with Anthropic (Claude) models only:

  • It activates a 1-hour prompt cache — cached input tokens cost 1/10 of the standard price.
  • Best for long, multi-turn complex tasks where you'll reuse the same context.
  • The first request costs 2x the standard input price (to create the cache), so short sessions may cost more with it enabled.
  • The default 5-13 minute cache already covers 90%+ of typical agent tasks.
Why is the model stuck on "thinking" for a long time?

Some models (e.g. DeepSeek R1) have extended reasoning modes that can take a while. Open Minis doesn't currently override model-level thinking behavior. If you prefer faster responses, switch to a non-reasoning model like Claude Sonnet or GPT-4o.

Skills

How do I install new skills?
  • Browse & install — go to Settings and browse the official skill repository.
  • Send a link — paste a GitHub URL of a SKILL.md file into a conversation, and the agent will install it.
  • Let the agent create one — describe what you need in plain language, and a capable model (e.g. Claude) can write and install a skill for you.
I installed a skill but its script files are missing?

Installing from a GitHub URL currently downloads only the SKILL.md file. If the skill depends on additional scripts (e.g. a scripts/ directory), ask the agent to download the full repository, or have it regenerate the scripts based on the SKILL.md description.

A skill I just created doesn't appear in Settings?

Skills created during a conversation may not immediately appear in the Settings page. Try leaving and re-entering the Settings screen to refresh.

Files & Storage

Are files isolated between conversations? How do I share files across sessions?
  • /var/minis/workspace/ — per-session, isolated between conversations.
  • /var/minis/memory/ — shared across all conversations.
  • /var/minis/skills/ — shared globally.

To share data between sessions, ask the agent to save it to the Memory or Skills directory.

How do I import files from the iOS Files app into a session?

Due to iOS sandboxing, the app's internal directories aren't visible in the Files app. Workarounds:

  • Paste file content into a conversation and ask the agent to save it.
  • Upload to iCloud Drive first, then ask the agent to read from the iCloud path.
Does "Clear Chat" delete workspace files?

Yes. Clearing a chat also removes that session's workspace files. If you have important files, ask the agent to copy them to the Memory or Skills directory before clearing.

Terminal & Linux Environment

What languages and tools does the built-in terminal support?
EnvironmentStatus
PythonFully supported. Use uv for dependency management.
Node.jsPartially supported. Small packages work fine; large packages may have issues.
Go / RustSupported.
FFmpegBuilt-in with hardware acceleration.
Do installed packages persist across conversations?

Yes. Packages installed via apk or pip persist in the Alpine Linux filesystem across all conversations, as long as you don't perform a Reset Rootfs.

The tar command hangs / GitHub CLI (gh) crashes?
  • tar hanging — the bundled BusyBox tar has a known issue. Run apk add tar to install the standalone version.
  • gh CLI — has compatibility issues under iSH. Use the GitHub API with a GITHUB_TOKEN environment variable as a workaround.

Automation & Background Tasks

How do I keep the agent running in the background?

Enable "Background Audio Keep-Alive" in Settings. This plays silent audio to prevent iOS from suspending the app. It stops automatically when the task finishes and doesn't interfere with other audio playback.

Can I set up scheduled / automated tasks?

Yes, via iOS Shortcuts. Open Minis provides Shortcuts actions — create an automation in the Shortcuts app and set it to run on a schedule. Note: cron inside the Linux environment won't persist due to iOS sandbox restrictions.

Data & Privacy

Where is my chat history stored?

All chat history is stored locally on your device. Nothing is uploaded to any cloud service. Switching your Apple ID does not affect local data.

Is multi-device sync supported?

Not yet. Conversations and workspace files are local only. iCloud sync is under development for future releases.

Other

How do I handle long conversations that become slow?

When context exceeds ~80k tokens (especially with images or large documents), rendering may lag. Solutions:

  • Ask the agent to summarize the conversation into a memory file.
  • Start a new session and instruct it to read the summary file to continue.
Where do I find alarms I created through the agent?

Alarms created via the agent won't appear in the iOS Clock app (system API limitation). View them by tapping the alarm icon on the Open Minis home screen, or ask the agent to list your alarms.

Can the agent open other iOS apps?

The agent can open apps that support URL Schemes (e.g. Maps, Settings, Safari). However, it cannot control or interact with other apps' UI due to iOS sandboxing.