Fast, streaming conversations
Pick any GGUF model — download from the built-in catalog or load a file from your device. Markdown, code blocks and history included.
Chat, see, listen and create — with models that run on your phone, laptop and browser. No account. No ads. Your prompts never leave your device unless you say so.
Get Llamatik
Start in a browser tab, then take it with you. Same chats, same models, same privacy.
Zero install. Runs in your browser with WebAssembly.
Open web app →Google Play · Android 8.0+
Get it on Google Play →App Store · iOS 16.6+
Download on the App Store →Apple Silicon & Intel · .dmg
Download for Mac →Windows 10+ · .msi
Download for Windows →Debian & Ubuntu · .deb · Omarchy
Download .deb →Every desktop release ships with SHA-256 checksums. All downloads & checksums →
What it does
Text, vision, voice, images, video and documents — each one running on the engine you already trust, on the hardware you already own.
Pick any GGUF model — download from the built-in catalog or load a file from your device. Markdown, code blocks and history included.
Ask about photos, screenshots and documents with on-device vision models like SmolVLM.
Whisper transcribes speech offline — with timestamps, translation to English, and speaker turns.
Describe a scene and get a short clip — encoded to H.264, saved to your gallery, with optional on-device upscaling.
Text-to-image and image-to-image, with live progress and ControlNet guidance for pose, depth and edges.
Planner, tools, permissions and an audit log — with companion personas from Professional to Pet.
Local RAG: chunking, embeddings and MMR retrieval — your files never get uploaded.
Speculative drafting for models that support it (Qwen3.5, GLM-4) — more tokens per second.
Models and chats stay on your device. No account, no ads, optional diagnostics that are off by default, and a clear warning before you ever connect to a server.
Share this device new in 2.0
Your gaming PC has the horsepower. Your phone has you. Turn on Share this device and Llamatik serves your loaded models to your other devices over the local network — found automatically with mDNS.
Connections between devices are currently plain HTTP — keep them on networks you trust.
Llamatik 2.0
The app is the showcase. The engine underneath is now a full Kotlin Multiplatform AI SDK.
Chat, models, downloads, RAG and the new Assistant facade — UI-free and publishable.
A Kotlin DSL and an AgentRuntime with permissions, confirmations and audit.
Text-to-video, ControlNet, live progress and video upscaling.
Multimodal image analysis through libmtmd — SmolVLM and friends.
Connect to servers, discover nodes, and share your own device.
Built on the new semantic-versioned releases of llama.cpp and whisper.cpp.
For developers
The same stack that powers the Llamatik app is on Maven Central. One Kotlin API for Android, iOS, Desktop and WASM — no Python, no glue code.
Click a module to copy its Gradle line.
val agent = Agent {
name("Kotlin Assistant")
systemPrompt("You are an expert Kotlin developer.")
temperature(0.7f)
memory(SlidingWindowMemory(windowSize = 20))
tools { builtIns() } // calculator, date/time, uuid, random
}
agent.stream("What is 17% of 2,340?").collect { chunk ->
when (chunk) {
is AgentStreamChunk.Delta -> print(chunk.text)
is AgentStreamChunk.Done -> println()
else -> Unit
}
}val path = LlamaBridge.getModelPath("phi-2.Q4_0.gguf")
LlamaBridge.updateGenerateParams(
temperature = 0.7f, maxTokens = 512,
topP = 0.95f, topK = 40, numThreads = 4,
)
LlamaBridge.initGenerateModel(path)
// Same call on Android, iOS, Desktop and WASM
val answer = LlamaBridge.generate(
"Explain Kotlin Multiplatform in one sentence."
)commonMain.dependencies {
// llama.cpp, whisper.cpp, stable-diffusion.cpp — one Kotlin API
implementation("com.llamatik:core:<version>")
// Chat, models, RAG, Agent DSL, Assistant facade
implementation("com.llamatik:sdk:<version>")
// AgentRuntime: planner, permissions, audit, workflows
implementation("com.llamatik:sdk-agent:<version>")
}$ curl -N localhost:8080/v1/generation/stream \
-H "Content-Type: application/json" \
-d '{"prompt":"Say hi from my own server"}'
Llamatik Server
Need more power than a phone? Self-host a Ktor server that speaks exactly the same API as the library — text, embeddings, speech, images, video and vision, streamed over SSE.
LlamaBridge, WhisperBridge, StableDiffusionBridge and MultimodalBridge paritycom.llamatik:server in your own appLlamatik Code
A privacy-first coding assistant for IntelliJ IDEA, Android Studio and the JetBrains family. Load a GGUF model once and your code never leaves your machine.

Why local-first
Prompts, documents and voice are processed on your hardware. There is no account to create and no cloud to trust.
On a plane, in a basement, behind a firewall. Once a model is downloaded, the network is optional.
MIT-licensed and built in the open. Inspect it, fork it, ship it. A 2026 Kotlin Foundation Grant winner.
What users say
“INCROYABLE !!!! MERCI AU DÉVELOPPEUR🤩 cette application gratuit illimitée sans soucis sans pub marche en hors ligne merci beaucoup je mettrai 10 étoiles si je pouvais”
“muito bom, cumpre oque promete! … Mas o app cumpre exatamente oque diz você consegue ter acesso a ia sem Internet.”
“love it, just like the library <3 thanks for your work”
“superb local integration functions on Oppo android device.”
Excerpts from public Google Play reviews, shown in the language they were written in and with first names only. They are a selection of the most positive ones; the full, unfiltered list is on the store page.
Compared
An honest look at Llamatik next to the tools people usually consider. Cloud models are more powerful; Llamatik trades some raw power for privacy, offline use and ownership.
| Feature | Llamatik | Ollama | LM Studio | llama.app | ChatGPT | Claude |
|---|---|---|---|---|---|---|
| Runs entirely on your device | Yes | Yes | Yes | Yes | No | No |
| Works offline | Yes | Yes | Yes | Yes | No | No |
| Native Android and iOS apps | Yes | No | No | No | Yes | Yes |
| Runs the model inside a browser tab | Yes | No | No | No | No | No |
| Vision, voice and image generation on device | Yes | Partly | Partly | Partly | No | No |
| Open source | Yes | Yes | Partly | Yes | No | No |
| Embeddable Kotlin Multiplatform SDK | Yes | No | No | No | No | No |
| Share one device's models with your other devices | Yes | Partly | Partly | No | No | No |
| No account required | Yes | Yes | Yes | Yes | Partly | No |
| Free to use | Yes | Yes | Yes | Yes | Partly | Partly |
| Frontier-scale model quality | No | Partly | Partly | Partly | Yes | Yes |
Based on public information as of November 2026. Products change quickly: if something is out of date, tell us at info@llamatik.com and we will fix it.
The Llamatik blog

Llamatik has been selected as a 2026 Kotlin Foundation Grant winner. Discover how this recognition will help advance on-device AI, Kotlin Multiplatform development, and local AI agents.

Llamatik Code is a local-first AI coding assistant for IntelliJ-based IDEs. All inference runs on your machine using GGUF models — no cloud, no API keys, no data leaving your device.

This article shows how to run LLMs offline on Android using Kotlin, with no servers, no API keys, and full user privacy-using Llamatik, a Kotlin-first wrapper around llama.cpp.
Ask your first question in seconds — no sign-up.