Llamatik Review: Free Open-Source Offline AI Chatbot for Android, iPhone and Desktop

Llamatik is an offline chatbot app that runs open-weight language models on the phone instead of calling a cloud service. The Google Play description calls it an experimental showcase of the Llamatik Kotlin Multiplatform library. No model is bundled: you download at least one inside the app, pick it in a model selector and chat. The listing names Gemma 3 270M, SmolVLM 256M and 500M, Phi-1.5, Qwen 2.5 and a quantized Llama 3.2 1B.

Three products share the name. This review covers the chat app, titled "Llamatik - Local AI Chatbot" on Google Play (package com.llamatik.app.android). The Llamatik SDK is the library that powers it, and Llamatik Server is a separate remote-inference backend from the same project. The website also lists a Llamatik Code plugin for JetBrains IDEs, which this review does not cover.

Llamatik is a download-and-open install on every platform it supports: two mobile stores plus desktop installers from the website. The links below are the ones the developer's website uses.

Setup follows the order described in the store listing; PromptQuorum has not run these steps.

Every item below comes from the Google Play description, the website or the repository README.

The app is the public demo of a larger open-source project. The repository at github.com/ferranpons/Llamatik is MIT-licensed, was created in July 2025 and shows 206 stars and 12 open issues on 19 October 2026. It holds the apps, the Kotlin Multiplatform SDK and the native bindings for llama.cpp, whisper.cpp and stable-diffusion.cpp.

According to the README, release 2.0.0 of 14 October 2026 renamed the inference library to com.llamatik:core and added a business-logic SDK with chat, model management and retrieval, an agent framework and an agent runtime with permission handling. These parts are aimed at developers building their own Android, iOS, desktop or WASM apps; they are separate from the chat app a reader installs from the store.

Chat content stays on the device by default, and the developer says nothing typed into the chatbot is sent to Firebase or any other server. The Google Play Data safety section declares no data shared with third parties. It lists crash logs, diagnostics and other app performance data as collected for app functionality and analytics, encrypted in transit, and it says the developer provides no way to request deletion of that data.

The privacy policy, last updated on 20 November 2025, names Google Analytics for Firebase and Firebase Crashlytics, and says Firebase Performance Monitoring may be enabled. Analytics events are pseudonymous (screen views, device type, operating system version, language and performance metrics). Crash reports carry the device model, operating system version, stack traces and app version, but no chat content. Model files come from public hosts such as Hugging Face, and the app does not log which model you choose.

The website says "No accounts. No tracking." That wording is narrower than the policy, because Firebase analytics and crash reporting still apply; the Google Play description itself says the internet is used only to download models and for analytics and crash reporting.

Llamatik is a free, MIT-licensed offline chatbot for Android, iPhone and desktop that runs small models through llama.cpp and shows what its developer's Kotlin SDK can do. It is open source and the repository was updated days before this review. Against that, it is an experimental demo with small models, a modest public rating, store builds that are five months old, and Firebase analytics that still leave the device. It suits readers who want an open-source way to try on-device chat, or who evaluate the SDK; readers who want more features can compare PocketPal AI or Private Mind.