LM Mini

Website · Desktop downloads · Docs

Chat with local and on-device AI from your phone, tablet, watch or desktop.

LM Mini is a Flutter chat client for LM Studio, Ollama, oMLX, Jan, Unsloth and other OpenAI-compatible servers, plus on-device models (llama.cpp and MLX). It runs on iOS, iPadOS, Android, macOS and Windows, with an Apple Watch companion.

Open core

This repository is an exported snapshot of the LM Mini app source. The free features are complete here and build from this tree.

LM Mini Pro implementations are not in this repository:

• lib/pro/ holds public stubs (line 1 of each file is // LM-MINI-PRO-STUB). They keep the signatures the app calls and do nothing else. ProFeatures.included is false in this build.
• packages/premium_stub/ is the no-op entitlement API (isPremium is always false).

So Pro entry points and upgrade prompts are hidden in this build. Pro is available in the official LM Mini app:

• App Store: https://apps.apple.com/app/lm-mini/id6751125309
• Google Play: https://play.google.com/store/apps/details?id=net.neuro9.lmmini

More about the open-source build: https://lmmini.com/open-source.html

What's in this repo

Everything below is free and builds from this tree.

Chat with your own models

• LM Studio, Ollama, oMLX, Jan, Unsloth and other OpenAI-compatible servers on your network
• Finds LM Studio on your Wi-Fi automatically
• USB mode: chat with your Mac over a cable, no network needed
• Custom headers and Cloudflare Access for servers behind a proxy
• API keys and tokens kept in the Keychain (iOS) and Keystore (Android)
• Clear connection status: offline, mobile data and unreachable-server messages, with tap-to-retry for unsent messages

On-device AI

• Six starter GGUF models (Qwen 3.5, Gemma 4, Bonsai 1-bit, LFM 2.5, MiniCPM 5) plus one imported GGUF (llama.cpp)
• Apple Intelligence on supported devices (iOS 26 / macOS 26), no download

Conversations

• Personas, including SillyTavern character card import (PNG, JSON, CHARX)
• Folders, themes, image and file attachments, and vision models
• Thinking / reasoning controls and global model parameters
• iCloud sync
• Export to PDF (Latin, Greek and Cyrillic), TXT and ZIP

Tools

• SearXNG web search
• MCP servers (LM Studio integrations and your own MCP URLs)

Voice

• Speech-to-text with Whisper
• System voices, plus on-device Kokoro (English, Spanish, French, Chinese) and Piper (German, Russian) voices
• Hands-free voice conversation

Image generation

• AUTOMATIC1111 and ComfyUI
• On-device Stable Diffusion with the starter model

Everywhere you are

• iPhone, iPad, Android, Mac and Windows
• Apple Watch companion
• Home-screen widgets
• Siri and Apple Shortcuts, including your LM Studio MCP tools
• Arena: race models on built-in prompts or your own, compare answers side by side, and keep a history of your races

Languages: English, German, Spanish, French, Russian and Chinese.

Build

You need the Flutter SDK (stable channel).

1. Firebase configuration. The app expects Firebase config files, which are not in the repository. Copy the placeholders, or run flutterfire configure against your own Firebase project (sign-in, sync and feedback only work with a real project):

   cp lib/firebase_options.example.dart lib/firebase_options.dart
   # iOS and macOS (the Xcode projects reference these files)
   cp ios/Runner/GoogleService-Info.example.plist ios/Runner/GoogleService-Info.plist
   cp macos/Runner/GoogleService-Info.example.plist macos/Runner/GoogleService-Info.plist
   # Android (optional: the Google Services plugin is applied only when
   # android/app/google-services.json exists)
   cp android/app/google-services.example.json android/app/google-services.json
   

   Google Sign-In is optional. To use it with your own project, replace the placeholder GIDClientID and the two CFBundleURLSchemes entries (com.googleusercontent.apps.… and app-1-…) in ios/Runner/Info.plist with the values from your GoogleService-Info.plist, and pass your web client ID on Android with --dart-define=GOOGLE_SIGN_IN_WEB_CLIENT_ID=<id>.apps.googleusercontent.com.

2. Fetch packages and check the tree:

   flutter pub get
   flutter analyze
   flutter test
   

3. Run it: flutter run (pick a device with -d).

   Native builds without running:

   # Android: works with or without android/app/google-services.json
   flutter build apk --debug
   # iOS Simulator: the Apple Watch companion target makes Flutter require a
   # simulator id (list them with `flutter devices` or
   # `xcrun simctl list devices available`)
   flutter build ios --simulator --debug --no-codesign -d <simulator-id>
   

   The first iOS or macOS build runs pod install, which can take several minutes. Device and macOS builds are signed: in Xcode, set the Runner (and, for iOS, the watch) targets to your own team and bundle identifier.

4. macOS desktop runtime (optional). The vendored llama.cpp server binaries are not in the repository; fetch them with:

   ./macos/scripts/vendor_llama_server.sh
   

Contributing

See CONTRIBUTING.md. Coding agents: see AGENTS.md. Release notes: CHANGELOG.md. Privacy policy: privacy_policy.html.

License

The code is MIT licensed, see LICENSE. The LM Mini name and app icon are not covered by that license, see TRADEMARK.md.

Special thanks

• @illava for reviewing and polishing the Chinese translation (#1).