LLM-Hub

What it is

LLM Hub is an open-source mobile app for on-device LLM chat, image generation, and video generation, available for both Android and iOS. It's optimized for mobile usage (CPU/GPU/NPU acceleration) and supports multiple model formats so you can run powerful models locally and privately.

How it works

The project integrates various runtimes and SDKs (e.g., Run Anywhere SDK, GeniesX, WhisperKit, Llama.cpp) to enable LLM inference, ASR, and media generation on-device. It supports RAG memory, web search, TTS auto-readout, and multimodal input for chat, plus image and video generation capabilities using Stable Diffusion and related tools. Detailed technical mappings are described in the Development Setup and Technology sections of the README.

Getting started

Development/setup steps included for Android and iOS:

• Android local development:

  git clone https://github.com/timmyy123/LLM-Hub.git
  cd LLM-Hub/android
  ./gradlew assembleDebug
  ./gradlew installDebug
  

• Android local configuration for Hugging Face token:

  HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
  

  DEBUG_PREMIUM=true
  

• iOS local development:

  git clone https://github.com/timmyy123/LLM-Hub.git
  cd LLM-Hub
  open ios/LLMHub/LLMHub.xcodeproj
  

  For signing and Xcode beta switches there are additional commands:

  sudo xcode-select -s /Applications/Xcode-beta.app/Contents/Developer
  xcodebuild -version
  

## Recent releases
Latest releases (ios-sdk-v0.19.7-patched-v16, v15, v14, v13, v12) with notes:
- ios-sdk-v0.19.7-patched-v16 (2026-06-03): Patched RunAnywhere SDK v0.19.7. UPDATE: iOS LlamaCPP build uses official ggml-org/llama.cpp b9493 (clean build, cache cleared, 2026-06-04 UTC). REVERT: gpu_layers override logic restored to origii
- ios-sdk-v0.19.7-patched-v15 (2026-05-20): Patched RunAnywhere SDK v0.19.7. UPDATE: iOS LlamaCPP build uses official ggml-org/llama.cpp b9247 (clean build, cache cleared, 2026-05-20 UTC). REVERT: gpu_layers override logic restored to orig
- ios-sdk-v0.19.7-patched-v14 (2026-05-20): Patched RunAnywhere SDK v0.19.7. UPDATE: iOS LlamaCPP build uses official ggml-org/llama.cpp b9247 (latest checked 2026-05-20 UTC). REVERT: gpu_layers override logic restored to original unconditi
- ios-sdk-v0.19.7-patched-v13 (2026-05-20): Patched RunAnywhere SDK v0.19.7. UPDATE: iOS LlamaCPP build uses official ggml-org/llama.cpp b9246 (latest checked 2026-05-20 UTC). REVERT: gpu_layers override logic restored to original unconditi
- ios-sdk-v0.19.7-patched-v12 (2026-05-20): Patched RunAnywhere SDK v0.19.7. UPDATE: iOS LlamaCPP build uses official ggml-org/llama.cpp b9246 (latest checked 2026-05-20 UTC). FIX: Ignore default -1 for user-provided gpu_layers to respect l

## Traction
Star count as of now: 507. No 1-day or 7-day star gain reported in the data provided (stars_1d: 0).

## Behind the repo
Not provided in the facts beyond license notes and open-source license: PolyForm Noncommercial License 1.0.0 for source code; commercial licensing inquiries via timmy@llm-hub.app.

## Caveats
License: PolyForm Noncommercial License 1.0.0 (non-commercial use allowed; commercial use requires explicit written permission). Age: created 2025-07-08; last_push 2026-07-21. Development notes include sensitive developer instructions (token handling, local config) and build guidance for Android and iOS.