LLamatikRun LLMs locally. Build in Kotlin. Stay in control.
Llamatik is a Kotlin-first, open-source library for running large language models on-device or remotely — powering private, offline-first AI apps across Android, iOS, desktop, and server.
No cloud lock-in. No data leakage. Just Kotlin + llama.cpp
Open source · Kotlin Multiplatform · On-device by default
Powered by proven technology
Built on battle-tested technologies used in production apps
Compose Multiplatform
Material Design 3
Why developers choose Llamatik
Llamatik removes the complexity of integrating LLMs into Kotlin apps.
You get a single, consistent API across platforms, native performance via llama.cpp, and full control over where inference runs — without rewriting your app or compromising on privacy.
A private, offline-first AI app — powered by Llamatik
The Llamatik app demonstrates what’s possible with on-device LLMs: fast, private AI chat running fully offline on your device. No accounts. No tracking. No cloud.
Everything you need to ship LLM features 🚀
Used in production apps across mobile and desktop
Llamatik focuses on the hard parts of integrating large language models — so you
can focus on building great products.
Instead of stitching together native code, bindings, and platform-specific logic,
Llamatik provides a unified Kotlin-first abstraction that works across platforms and
environments.
On-device inference
Run LLMs fully offline using native llama.cpp bindings compiled for each platform. No network required, no data leakage.
Kotlin Multiplatform API
A single, shared Kotlin API for Android, iOS, desktop, and server — with expect/actual handled for you.
Remote inference
Use HTTP-based inference when models are too large or when centralized execution is required — without changing your app logic.
Text generation & embeddings
Built-in support for common LLM use cases like text generation, chat-style prompts, and vector embeddings.
GGUF model support
Works with modern GGUF-based models such as LLaMA, Mistral, and Phi.
Lightweight runtime
No heavy frameworks, no cloud SDKs. Just Kotlin, native binaries, and full control over your stack.
Why run LLMs on-device?
Local-first beats cloud-first when it matters
Cloud-based LLMs are powerful — but they come with trade-offs. Latency, recurring costs, privacy concerns, and vendor lock-in can quickly become blockers for real-world applications.
Llamatik lets you run models directly on user devices using native inference powered by llama.cpp, reducing infrastructure costs while keeping sensitive data local. When you need scale or centralized inference, you can seamlessly switch to remote execution — using the same Kotlin API.
Local-first by default. Remote when it makes sense. This reframes cost as:
- 💸 Lower infrastructure costs
- 🔒 Privacy by default
- 🧠 Simpler architecture
Open source. Developer-first.
Llamatik is fully open source and built in the open.
No closed binaries. No hidden services. Just code you can inspect, run, and extend.
View on GitHub
