Llamatik
On-device AI SDK for Kotlin Multiplatform running LLMs, STT, and diffusion.
Run llama.cpp, whisper.cpp, and stable-diffusion.cpp natively on Android, iOS, Desktop, and WASM.
Llamatik is a Kotlin Multiplatform SDK that brings on-device AI inference, agent frameworks, and RAG to mobile and desktop apps. It suits Kotlin developers building offline-first applications that need local LLMs, speech-to-text, or image generation without Python servers.
What it is
The SDK combines direct bindings to native C++ inference engines like llama.cpp and whisper.cpp with a high-level Agent Framework DSL. It provides an injectable Assistant facade, persistent vector stores, workflow engines, and built-in permission management for platform tools.
Who it's for
It suits mobile and desktop software engineers writing Kotlin who need privacy-first, offline AI capabilities inside native apps. It does not suit developers working outside the Kotlin ecosystem or those looking for ready-made cloud APIs.
Key features
Kotlin Multiplatform SDK
Share AI business logic across Android, iOS, Desktop, and WASM using a unified Kotlin API.
Agent Framework DSL
Build agentic workflows with memory types, built-in tools, streaming output, and approval gates.
Native C++ Engines
Execute local text generation, multimodal vision, speech-to-text, and image generation via optimized backends.
Agent Platform Runtime
Manage permissions, persistent memory, audit logs, and companion profiles with 9 built-in platform tools.
Up and running in 3 steps
Our honest verdict
What's great
- Runs completely on-device with zero server dependency or network requirement.
- Combines low-level C++ inference bindings with a high-level Kotlin agent DSL.
- Supports multi-token prediction and schema-constrained JSON output generation.
- Includes multimodal vision, speech-to-text, and stable diffusion capabilities.
Where it falls short
- Requires writing Kotlin and managing native mobile or desktop compilation targets.
- On-device inference is constrained by the hardware memory and compute limits of the host device.
- Running large language or diffusion models locally consumes significant battery and storage resources.
Pricing & license
The repository does not state a license. The only cost is running the software yourself on your own hardware.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
What platforms are supported by Llamatik?
Llamatik supports Android, iOS, Desktop (JVM), and WASM through Kotlin Multiplatform.
Does Llamatik require an internet connection?
No, it is offline-first and runs models entirely on-device using native C++ inference engines.
Can I use Llamatik for speech-to-text and image generation?
Yes, it includes built-in bindings for whisper.cpp for transcription and stable-diffusion.cpp for image and video generation.
