Archon AI 🤖

Offline Android AI Assistant (llama.cpp · GGUF · Kotlin · Jetpack Compose)

Run state-of-the-art LLMs fully offline on Android using llama.cpp and GGUF models.
No cloud. No servers. No tracking. Everything runs locally on-device.

This project demonstrates that multiple major model architectures run flawlessly on Android when integrated with llama.cpp, Kotlin, and Jetpack Compose.

---

📱 App Availability

The app is available as a pre-built APK here:

🔗 Download Archon AI (APK)
https://github.com/AnishKMBtech/archon-ai-Llama.cpp-kotlin-integration/releases/tag/master

ℹ️ The source code is not uploaded.
This repository focuses on architecture, implementation, and proof-of-work, not cloning or building.

---

🖼️ Screenshots

LOGO Preview

More Screenshots

---

✨ Key Features

• 💬 Offline Chat with LLMs
• 🔒 100% Private – No Internet Required
• ⚡ Real-time Token Streaming
• 🔁 Model Loading & Switching
• 📥 Local GGUF Model Support
• 🧠 Multiple Model Architectures Supported
• 🎨 Modern UI with Jetpack Compose
• 📱 Optimized for Mobile Devices

---

🧠 GGUF Models Tested (llama.cpp)

The following models were tested successfully and run flawlessly in Archon AI:

• ✅ DeepSeek 1.5B
• ✅ Qwen 2.5 – 0.5B
• ✅ Gemma 1B Instruct
• ✅ Llama 3 – 1B Instruct

➡️ This confirms that major architectures (LLaMA, Qwen, Gemma, DeepSeek) are fully compatible with llama.cpp on Android.

⚠️ Performance depends on your device RAM and storage.

---

🤗 GGUF Models (Ready to Use)

All tested and optimized GGUF models are available here:

🔗 Archon AI GGUF Models (Hugging Face)
https://huggingface.co/ANISH-j/gguf-models-for-archon-ai/tree/main

Recommended Quantization:

• Q4_K_M
• Q4_0

These offer the best balance for mobile performance vs quality.

---

📦 How Model Handling Works

📥 Model Download

• Download GGUF models manually (Hugging Face or local storage)
• Models are stored locally on-device

🔄 Model Loading

• Select a GGUF file from storage
• llama.cpp initializes the model
• Metadata and readiness are shown in the UI

🔁 Model Switching

• Switch between models without reinstalling the app
• Each model runs independently

⚡ Inference

• Token-by-token streaming
• Low-latency responses
• Optimized for ARM64 devices

---

🛠️ Tech Stack

• Inference Engine: llama.cpp
• Model Format: GGUF
• Language: Kotlin
• UI: Jetpack Compose
• Architecture: On-device / Edge AI
• Native Layer: C/C++ via JNI
• Target Devices: ARM64 Android phones

---

📱 Device Requirements

• Android: 13+
• RAM:
  • Minimum: 4 GB
  • Recommended: 6–8 GB+
• Storage:
  • 2–8 GB free (depends on model size)

---

🎯 Use Cases

• 🧑‍🎓 Learning & Q&A
• ✍️ Writing Assistance
• 💡 Idea Brainstorming
• 🌍 Language Practice
• 💻 Code Explanations
• 🔐 Fully Private AI Assistant

---

⚠️ Notes & Limitations

• Large models may cause OOM crashes on low-RAM devices
• Thermal throttling may affect long sessions
• Smaller models (≤1.5B) are ideal for phones

---

🙏 Credits

• llama.cpp – https://github.com/ggerganov/llama.cpp
  The backbone of on-device LLM inference

• GGML community & open-source contributors

---

📜 License

This project integrates llama.cpp, licensed under the MIT License.
Model licenses depend on their respective authors.

---

🔒 Archon AI

Your personal, offline, edge AI assistant — powered by llama.cpp