Axiom AI - On-Device AI Runtime for Android

Run ChatGPT-like features fully offline on Android in minutes. No APIs. No cloud. No data leakage.

---

✨ Why Axiom AI?

Most AI SDKs require:

• ❌ Cloud APIs
• ❌ Latency
• ❌ Privacy tradeoffs
• ❌ Expensive scaling

Axiom AI flips that:

• ✅ 100% on-device inference (powered by llama.cpp)
• ✅ Offline-first (works in airplane mode)
• ✅ Kotlin-first clean API
• ✅ Play Store–safe model delivery
• ✅ Production-ready architecture

---

📦 What You Get

• 🧠 Local LLM inference (GGUF models)
• ⚡ Streaming text generation
• 📥 Built-in model download manager
• 🧩 Modular architecture
• 📱 Android-optimized performance

---

🎥 Demo

Offline AI running on a real Android device (no internet)

Add demo GIF/video here

---

⚡ Quick Start (1 Minute)

1. Add dependencies

dependencies {
    implementation("com.axiom:axiom-core:0.1.0")
    implementation("com.axiom:axiom-llama-cpp:0.1.0")
    implementation("com.axiom:axiom-models:0.1.0")
}

---

2. Initialize engine

val engine = LlamaCppEngine()
val modelManager = DefaultModelManager(context)

// Get recommended model
val model = modelManager.recommend(context)

// Download if needed
val modelPath = modelManager.download(model) { progress ->
    Log.d("Axiom", "Downloading: $progress")
}

// Initialize engine
engine.init(LLMConfig(
    modelPath = modelPath,
    contextSize = 1024,
    temperature = 0.7f,
    topK = 40
))

---

3. Generate text (streaming)

engine.stream("Explain Android in simple terms") { token ->
    print(token)
}

---

Alternative: Manual Model Setup

For the sample app, you can manually download and add a model:

1. Download TinyLlama 1.1B (Q4_K_M):

   wget https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf

2. Add to sample app assets:

   mkdir -p sample/src/main/assets/models
   cp tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf sample/src/main/assets/models/tinyllama.gguf

3. Rebuild sample app:

   ./gradlew :sample:assembleDebug

---

🧠 Architecture

axiom-core         → SDK interfaces + config
axiom-llama-cpp    → llama.cpp engine binding
axiom-models       → model manager + downloader
sample-app         → demo implementation

---

📥 Model Delivery (Play Store Safe)

Axiom does NOT bundle models in APK.

Instead:

• Models are fetched from a remote registry
• Downloaded via Android DownloadManager
• Stored in app-private storage
• Verified via checksum

✅ No APK bloat ✅ No policy violations ✅ Resumable downloads

---

📚 Supported Models

Model	Size	RAM	Speed
TinyLlama 1.1B	~500MB	3–4GB	⚡ Fast
TinyMistral 0.2B	~130MB	2–3GB	⚡⚡ Very Fast
Mistral 7B	~4GB	6–8GB	🐢 Slow but high quality

---

🧩 API Overview

Engine

interface LLMEngine {
    suspend fun init(config: LLMConfig): Boolean
    suspend fun generate(prompt: String): String
    suspend fun stream(prompt: String, onToken: (String) -> Unit)
    fun cleanup()
}

---

Config

data class LLMConfig(
    val modelPath: String,
    val contextSize: Int = 1024,
    val temperature: Float = 0.7f,
    val topK: Int = 40,
    val threads: Int = Runtime.getRuntime().availableProcessors()
)

---

Model Manager

ModelManager.fetchRegistry()

ModelManager.download(model) { progress -> }

ModelManager.getInstalledModels()

ModelManager.delete(model)

---

⚙️ Performance Tips

• Use Q4_K_M quantization
• Keep context ≤ 1024 for mobile
• Prefer small models for UX
• Avoid frequent init/cleanup

---

🔐 Privacy

• All inference runs locally
• No data leaves device
• No API calls

---

🛣 Roadmap

v0.2

• Streaming API improvements
• Model auto-selection

v0.3

• GPU / NNAPI acceleration
• Background downloads

v1.0

• Stable SDK
• Production benchmarks
• UI components (ChatKit)

---

🤝 Contributing

We welcome contributions!

Good first issues:

• Add new model configs
• Improve streaming API
• Optimize memory usage
• Enhance sample app UI

Setup

git clone https://github.com/av-feaster/axiom
cd axiom
./gradlew build

---

📜 License

MIT License

---

💬 Vision

Axiom AI aims to become:

"Firebase for on-device AI"

---

⭐ Support

If this project helps you:

• ⭐ Star the repo
• 🧵 Share on Twitter / Reddit
• 🧪 Build something cool

---

🧠 Final Thought

The future of AI is not just in the cloud.

It's on your device.

---

Made with ❤️ for indie developers