v1.0.0 Release Available

Offy: Offline AI Companion — Your LLM, entirely on your phone.

No internet. No cloud. No API keys. Just you and a language model running in your pocket. Because your conversations should stay yours.

Download APK
Free forever. For Android 11+.

Direct download secure GGUF chat assistant • SHA-256 Verified

$flutter build apk --release
Offy App UI Screenshot

Why choose local AI?

Traditional AI clients stream your queries to massive cloud server arrays. Offy processes language locally on your handset CPU, shifting control back to your device.

Cloud AI

  • ✗Needs constant internet connection
  • ✗Chat logs saved and parsed on external servers
  • ✗Requires accounts, subscriptions, and API keys
  • ✗Pay-per-token pricing model
  • ✗Constant risk of data leaks and policy shifts

Offy

RECOMMENDED
  • ✓Works 100% offline (No connection needed)
  • ✓All parameters and weights load locally on-device
  • ✓Zero accounts, credentials, or keys required
  • ✓Free forever (Download once, query infinitely)
  • ✓Zero telemetry — complete architectural privacy

Key Features

Built from scratch for local Android inference, Offy delivers full chatbot capability with zero backend reliance.

Fully Offline LLM Chat

Run optimized localized language models entirely on-device with zero server endpoints or packet transmissions.

Complete Privacy

Your conversations are stored in a secure local database that never leaves your sandboxed memory stack.

Streaming Token Output

See generated characters stream in real-time. Watch instant performance logs like token counts and speeds.

Model Catalog & Import

Browse HuggingFace files inside the app UI download manager, or sideload customized GGUFs straight from storage.

No API Keys

No developer portals, zero configuration. Simply download your preferred model once and use it indefinitely.

Device Profiling

Auto-detects available device CPU core configurations and RAM boundaries to configure optimal threads and context.

SYSTEM ARCHITECTURE

Optimized Mobile CPU Inference

Offy bridges Dart FFI to llama.cpp source codes via a JNI layer. The inference kernel utilizes ARM64 Neon vector registers directly, bypassing translation runtimes.

ENGINE INFERENCE BACKEND
Llama_flutter_android (C++ via FFI)
SUPPORTED ARCHITECTURES
arm64-v8a Only (Android 11+)

Model Compatibility (Quantized GGUF formats)

Qwen 2.5 (1.5B / 3B / 7B)Excellent multi-lingual and logic performance
4-Bit QuantizedRecommended: 1.5B (4GB RAM)
Phi 3.5 Mini (3.8B)State-of-the-art math and code reasoning capability
Q4_K_M FormatRecommended: 6GB RAM
Mistral (7B)Balanced parameters for conversational responses
Q3_K_L / Q4_0Recommended: 8GB+ RAM
Gemma 2 (2B / 9B)Google-trained lightweight models tuned for mobile
Q4_K_S QuantRecommended: 6GB RAM
Users can download models directly inside the App from HuggingFace repositories.

Take control of your AI.Download Offy today.

Stop sending private chat history, location queries, and context notes to cloud models. Download the standalone APK client.