Home

llama-android

Run LLMs on Android with one Gradle line. A prebuilt llama.cpp AAR with a clean Kotlin coroutine API — chat completion and embeddings, fully on-device. No cloud, no API key, no per-token billing. Your users' conversations never leave the phone.

• No NDK / CMake / Python — just add the dependency.
• GGUF models — any quantization (Q4_0, Q4_K_M, Q5_K_M, …).
• CPU / NEON on arm64-v8a, 16 KB-page ready (Android 15).
• Wraps llama.cpp build b9878.

Start here

Page	What it covers
Installation	Add the AAR (Maven Central or JitPack).
Quick Start	Your first chat completion, step by step.
Choosing a Model	Which GGUF to ship, RAM budget, where to download.
Voice Assistant Pipeline	Whisper → llama.cpp → TTS, all on-device.
FAQ	Common questions.
Troubleshooting	Crashes, slow generation, UnsatisfiedLinkError.

The whole API in 6 lines

val model  = Llama.loadModel("/path/to/model.gguf")
val answer = Llama.complete(model, "Explain gravity to a 5-year-old.")
val vector = Llama.embed(model, "on-device AI")
val info   = Llama.getSystemInfo()
Llama.releaseModel(model)

That's the entire Free surface: loadModel, complete, embed, getSystemInfo, releaseModel. Need streaming, Vulkan GPU, JSON/grammar-constrained output, vision, or multiple concurrent sessions? See Free vs Pro.

---

Maintained by Jokobee · contact@jokobee.com