Is LLM Hub open-source?

Yes, LLM Hub is completely open-source. Download it from Google Play, the Apple App Store, or build it directly from source on GitHub.

Does LLM Hub work offline?

Yes. Internet is only required for the initial model download. After that, all features—AI chat, image generator, translator, transcriber, and coder—work 100% offline.

What AI models does LLM Hub support?

LLM Hub supports 15+ models, including Gemma-3, Gemma-4, Gemma-3n (multimodal), Llama-3.2 (GGUF), Phi-4 Mini, IBM Granite 4.0, LiquidAI LFM, Ministral-3, Whisper ASR, and Kokoro TTS.

Is my data private and secure?

Absolutely. All AI processing runs entirely on your local device. We collect zero data, require no accounts, and track no telemetry. Your privacy is 100% guaranteed.

What are the minimum requirements?

Android 8.0+ or iOS 17.0+, with a minimum of 2GB RAM (6GB+ recommended, 8GB+ for GPU acceleration) and 1-5GB storage space depending on the downloaded models.

What features does LLM Hub offer?

11 core offline AI tools: AI Chat, Writing Aid, Stable Diffusion Image Generator, Video Generator, Image Upscaler, Offline Translator, Voice Transcriber, Scam Detector, Vibes Coder sandbox, creAItors, and VibeVoice.

Does LLM Hub support vision and image input?

Yes. Gemma-3, Gemma-3n, and Ministral-3 support image and vision understanding natively on-device.

Does LLM Hub support text-to-speech?

Yes, LLM Hub features TTS auto-readout, which reads AI responses aloud for hands-free conversations.

Can I code apps on my phone offline?

Vibes is an on-device coding sandbox. Describe what you want, and the local AI generates running HTML/JS/CSS that renders instantly inside a secure local sandbox.

Can I create custom AI personas on-device?

creAItors is an offline AI persona builder. Customize unique system prompts and personality profiles to create your own specialized AI agents.

Is there an offline voice chat AI app?

VibeVoice is a real-time, hands-free voice chat mode powered by Whisper ASR and Kokoro TTS. Speak naturally and the local model responds instantly with real-time text-to-speech feedback.

Which phones run local AI fastest?

Yes, LLM Hub leverages native Qualcomm QNN frameworks to accelerate inference on Snapdragon NPU chips, enabling blazing fast offline speeds.

Is LLM Hub open-source?

Yes, LLM Hub is completely open-source. Download it from Google Play, the Apple App Store, or build it directly from source on GitHub.

Does LLM Hub work offline?

Yes. Internet is only required for the initial model download. After that, all features—AI chat, image generator, translator, transcriber, and coder—work 100% offline.

What AI models does LLM Hub support?

LLM Hub supports 15+ models, including Gemma-3, Gemma-4, Gemma-3n (multimodal), Llama-3.2 (GGUF), Phi-4 Mini, IBM Granite 4.0, LiquidAI LFM, Ministral-3, Whisper ASR, and Kokoro TTS.

Is my data private and secure?

Absolutely. All AI processing runs entirely on your local device. We collect zero data, require no accounts, and track no telemetry. Your privacy is 100% guaranteed.

What are the minimum requirements?

Android 8.0+ or iOS 17.0+, with a minimum of 2GB RAM (6GB+ recommended, 8GB+ for GPU acceleration) and 1-5GB storage space depending on the downloaded models.

What features does LLM Hub offer?

We offer 11 core offline AI tools: AI Chat, Writing Aid, Stable Diffusion Image Generator, Video Generator, Image Upscaler, Offline Translator, Voice Transcriber, Scam Detector, Vibes Coder sandbox, creAItors, and VibeVoice.

Does LLM Hub support vision and image input?

Yes. Gemma-3, Gemma-3n, and Ministral-3 support image and vision understanding natively on-device.

Does LLM Hub support text-to-speech?

Yes, LLM Hub features TTS auto-readout, which reads AI responses aloud for hands-free conversations.

Can I code apps on my phone offline?

Vibes is an on-device coding sandbox. Describe what you want, and the local AI generates running HTML/JS/CSS that renders instantly inside a secure local sandbox.

Can I create custom AI personas on-device?

creAItors is an offline AI persona builder. Customize unique system prompts and personality profiles to create your own specialized AI agents.

Is there an offline voice chat AI app?

VibeVoice is a real-time, hands-free voice chat mode powered by Whisper ASR and Kokoro TTS. Speak naturally and the local model responds instantly with real-time text-to-speech feedback.

Which phones run local AI fastest?

Yes, LLM Hub leverages native Qualcomm QNN frameworks to accelerate inference on Snapdragon NPU chips, enabling blazing fast offline speeds.

SEC.00//LLM-HUB.SYS // PRIVATE AI UNIT

AI that works
in airplane mode.

15+ AI models running 100% on your phone. Chat, images, video, music, code — with zero signal, zero accounts, zero tracking.

AIRPLANE MODE

NETWORKLINK: ONLINE

COREON-DEVICE CORE: RUNNING

LIVE SPECIMEN // TOKEN STREAM0 tok/s

$ write a haiku about dead zones

▊

simulated token stream — inference happens on your device

GITHUB

SEC.00B // PACKET INSPECTOR

Every outbound request. Blocked.

Watch the app try to phone home. There is nothing to phone home to.

0 TRACKERS/0 ACCOUNTS/0 BYTES SENT

OUTBOUND FIREWALL // LIVE

▊

SEC.02//FIELD TESTS

01Featured Tool: Image Generator

Generate Stunning Images, Fully On-Device

Experience the magic of Stable Diffusion running entirely offline on your phone. Powered by native NPU and GPU compilers for lightning-fast edge inference.

01.1

100% Private Offline Generation

Prompts, seeds, and generated images stay strictly on your local device storage. Zero external server requests.

01.2

Qualcomm NPU / GPU Accelerated

Optimized for mobile accelerators (CPU/GPU/NPU) utilizing high-speed MNN and QNN backends for Snapdragon chips.

01.3

Absolute Reality SD1.5 Checkpoint

Generates beautiful, photorealistic, and highly detailed art offline with high-fidelity, resource-efficient weights.

01.4

Zero Subscriptions, Infinite Art

No daily token limits, no wait queues, and no premium paywalls. Generate as many images as you like, whenever you want.

Stable Diffusion 1.5GPU/NPU EnabledNo server needed

REC ● OFFLINE

02Featured Tool: Vibes Coder

Design Apps and Sites, Preview in Real-Time

A full-featured mobile coding environment powered by local models. Write natural language descriptions and see dynamic, functional web views render before your eyes—no cloud server needed.

02.1

Instant HTML/JS/CSS Previews

Write prompts, generate full front-end code, and run it in a secure local sandbox with real-time browser previewing.

02.2

100% On-Device Code Models

Harness local coding intelligence powered by state-of-the-art quantized model weights running natively on your hardware.

02.3

Pure Local Playground

Zero external dependencies, zero cellular data consumed, and absolute privacy. Code and execute with 100% offline security.

02.4

Rapid Iterative Prototyping

Tweak details on the fly. Prompt for additions (e.g. 'add a toggle switch' or 'change to dark mode') and see updates instantly.

HTML/CSS/JS SandboxOn-Device LLM CoderOffline Preview

REC ● OFFLINE

03Featured Tool: AI Agent

Run Termux Commands, Control Apps & Maps

An autonomous on-device AI agent for Android & iOS. Generate and execute shell commands inside Termux, render location maps, compose messages, and update calendar schedules—100% offline.

03.1

Termux Command Execution

Generate CLI scripts and execute shell commands directly inside the app terminal environment.

03.2

Interactive Location & Maps

Search locations, view coordinates, and render interactive map views dynamically on demand.

03.3

Messaging & SMS Automation

Compose and dispatch text messages and notifications autonomously through local Android APIs.

03.4

On-Device Calendar Management

Create, search, and update calendar entries and reminders without external server calls.

Termux Shell RunnerInteractive MapsSMS & MessagingCalendar Automation

REC ● OFFLINE

04Featured Tool: Music Generator

Generate Music On-Device, Fully Offline

Run Google Magenta Realtime 2 on iOS or Stable Audio Small on Android and generate music on device.

04.1

Google Magenta Realtime 2 (iOS)

Generate expressive real-time audio and musical compositions directly on Apple silicon.

04.2

Stable Audio Small (Android)

Ultra-fast sound synthesis and high-quality audio generation optimized for mobile devices.

04.3

100% On-Device & Private

All music synthesis runs offline. Zero server requests, total privacy for your creative audio.

04.4

Real-Time Audio Synthesis

Generate high-quality music tracks instantly with GPU acceleration.

Google Magenta Realtime 2Stable Audio Small100% On-Device

REC ● OFFLINE

SEC.03//MODEL MANIFEST

Download once. Run forever.

15+ models, quantized for your pocket. Pick your weights, keep them on-device.

Gemma-3529MB

Llama-3.21.2GB

Phi-4 Mini3.67GB

IBM Granite 4.07B

LiquidAI LFM1.2B

Ministral-33B

Whisper ASRLIGHTWEIGHT

Kokoro TTSLIGHTWEIGHT

Gemma-3529MB

Llama-3.21.2GB

Phi-4 Mini3.67GB

IBM Granite 4.07B

LiquidAI LFM1.2B

Ministral-33B

Whisper ASRLIGHTWEIGHT

Kokoro TTSLIGHTWEIGHT

MODEL	PROVIDER	SIZE	CAPABILITY
Gemma-3 1B Optimized text models for mobile with GPU acceleration. Multiple quantizations and context windows (1.2k-4k)	GOOGLE	INT4 - 529MB · INT8 - 1005MB-1024MB	TEXT
Gemma-3 GGUF Multimodal GGUF multimodal models with vision projectors for CPU/GPU/NPU acceleration	GOOGLE	4B (3.0GB) · 12B (7.7GB)	VISION
Gemma-3n Multimodal Multimodal models with text, vision, and audio capabilities. Selective parameter activation	GOOGLE	E2B (3.15GB) · E4B (4.33GB)	VISION
Llama-3.2 (LiteRT) Meta's INT8 models for on-device inference with 4k context	META	1B (2.01GB) · 3B (5.11GB)	TEXT
Llama-3.2 GGUF GGUF format Llama models with 128k context and CPU/GPU/NPU support	META	1B & 3B - Multiple quantizations	TEXT
IBM Granite 4.0 H-Tiny Compact enterprise model with 128k context window and multiple quantizations	IBM	7B - Q2_K to f16 (2.59GB-13.9GB)	TEXT
IBM Granite 4.0 H-Small High-quality enterprise model with 128k context window	IBM	32B - Q2_K to f16 (11.8GB-64.4GB)	TEXT
Phi-4 Mini Microsoft's efficient model for advanced reasoning with GPU support on 8GB+ devices	MICROSOFT	INT8 (3.91GB)	TEXT
LFM-2.5 1.2B Instruct LiquidAI's efficient instruction model with 128k context. Available in GGUF and ONNX formats	LIQUIDAI	GGUF & ONNX variants	TEXT
LFM-2.5 1.2B Thinking Reasoning model with 'thinking' mode support and 128k context. Available in GGUF and ONNX	LIQUIDAI	GGUF & ONNX variants	TEXT
LFM-2.5 VL 1.6B Vision-language model supporting both text and image input with 128k context	LIQUIDAI	GGUF with vision projectors	VISION
Ministral-3 3B Multimodal instruction model with vision support and 262k context. Available in GGUF and ONNX	MISTRALAI	GGUF & ONNX variants	VISION
Absolute Reality SD1.5 Image generation model for creating images from text prompts	STABLE DIFFUSION	MNN (CPU) · QNN (NPU)	IMAGE
Gecko-110M Compact embedding model for RAG memory system with multiple dimension options	GOOGLE	64D-1024D embeddings	RAG
EmbeddingGemma-300M High-quality text embeddings for enhanced RAG retrieval	GOOGLE	High-quality embeddings	RAG
Whisper ASR State-of-the-art automatic speech recognition (ASR) model for converting voice/audio to text offline	OPENAI	Highly lightweight quantizations	TEXT
Kokoro TTS State-of-the-art text-to-speech (TTS) model for producing highly natural voice synthesis on-device	KOKORO	Extremely lightweight & efficient	TEXT

Gemma-3 1B

GOOGLE

Optimized text models for mobile with GPU acceleration. Multiple quantizations and context windows (1.2k-4k)

INT4 - 529MB · INT8 - 1005MB-1024MBTEXT

Gemma-3 GGUF Multimodal

GOOGLE

GGUF multimodal models with vision projectors for CPU/GPU/NPU acceleration

4B (3.0GB) · 12B (7.7GB)VISION

Gemma-3n Multimodal

GOOGLE

Multimodal models with text, vision, and audio capabilities. Selective parameter activation

E2B (3.15GB) · E4B (4.33GB)VISION

Llama-3.2 (LiteRT)

12 tools. Zero signal required.

Every tool below runs fully offline. Pick your dead zone.

SCENARIO A//TUNNEL

AI Chat

Multi-turn conversations with 15+ AI models. RAG memory, web search, and TTS auto-readout.

Transcriber

Convert voice or audio to text offline using Whisper ASR and Gemma models.

Scam Detector

Analyze text, images, or URLs for potential scams with AI risk assessment.

SCENARIO B//AIRPLANE

Translator

Translate text, images (OCR), and audio across 50+ languages. Works fully offline.

Image Generator

Generate images from text prompts with Stable Diffusion — GPU/NPU accelerated.

Video Generator

Create AI videos from text and image prompts on-device — fully offline.

SCENARIO C//BASEMENT

Vibes Coder

Full on-device coding environment. Write a prompt, preview running HTML/JS/CSS instantly in a sandbox.

Writing Aid

Grammar, style, paraphrasing, tone adjustment, and code generation — all on-device.

creAItors

Create powerful custom AI personas — from pirate narrators to full-stack system architects.

SCENARIO D//ROAMING

VibeVoice

Hands-free AI voice chat with natural conversations, powered by Whisper ASR and Kokoro TTS for real-time voice input/output.

AI Agent

Autonomous on-device agent to run Termux shell commands, render interactive maps, send messages, and manage calendar events.

Music Generator

Run Google Magenta Realtime 2 on iOS or Stable Audio Small on Android and generate music on device.

Image Upscaler

Upscale and enhance images on-device — fully offline.

▸ 100% PRIVATE▸ NO INTERNET REQUIRED▸ ZERO DATA COLLECTION▸ OPEN SOURCE▸ GPU/NPU ACCELERATED

SEC.06//BOOT SEQUENCE

From download to offline in three commands.

Grab the weights once. Everything after that happens on your hardware.

llmhub — zsh

$ llmhub pull gemma-3n-e4b

# download .task weights from HuggingFace — once

$ llmhub load --backend litert

# load into MediaPipe LLM Inference API

$ llmhub chat

# local inference on CPU/GPU — 0 bytes leave the device

$ ▊

SEC.08//CASE STUDY

Featured Case Study | Y Combinator Backed

Featured by RunAnywhere:
Android to iOS in Six Weeks

Read the official case study by YC-backed RunAnywhere on how LLM Hub brought 15+ 100% on-device AI models to iOS with high-speed local inference.

YC BackedRunAnywhere SDK6 Weeks to iOS

Read Case Study on RunAnywhere

SEC.09//FAQ

FAQ

Frequently Asked Questions

Everything you need to know about LLM Hub, running 100% privately on-device.

SEC.10//TRANSMIT

Start Using Private AI Today

Open source. No accounts. No tracking. Just powerful AI running entirely on your Android or iOS device.

GITHUB

NO ACCOUNTS · NO TRACKING · OPEN SOURCE

AI that worksin airplane mode.

Every outbound request. Blocked.

Generate Stunning Images, Fully On-Device

Design Apps and Sites, Preview in Real-Time

Run Termux Commands, Control Apps & Maps

Generate Music On-Device, Fully Offline

Download once. Run forever.

12 tools. Zero signal required.

From download to offline in three commands.

Featured by RunAnywhere:Android to iOS in Six Weeks

Frequently Asked Questions

Start Using Private AI Today

AI that works
in airplane mode.

Featured by RunAnywhere:
Android to iOS in Six Weeks