Roost is an Android application developed by RinWave that enables users to run local AI models directly on their devices. Its primary function is to facilitate offline chat and model inference without requiring a constant internet connection, relying instead on locally stored machine learning models. The app also provides integration with Hugging Face, allowing users to discover and browse available models from that repository. Key capabilities include device-aware recommendations that suggest models optimized for the user’s specific hardware, as well as resumable downloads to handle large model files efficiently. Users can select from different backends for running models, and the app supports workflows that are compatible with local OpenAI and Ollama APIs, making it usable for developers who want to experiment with or deploy AI models in a self-hosted environment. Typical use cases involve running conversational AI or other generative models on Android without relying on cloud services, which can be useful for privacy, offline scenarios, or reducing latency. The workflow generally consists of discovering a model through Hugging Face, downloading it with the option to pause and resume, and then selecting an appropriate backend to run the model locally. The app is designed for users who have some familiarity with AI models and want to manage them directly on their mobile device. Technical requirements include an Android device capable of running local AI models. The app’s backend selection and device-aware recommendations help match models to the available computational resources. By supporting local OpenAI and Ollama compatible workflows, Roost allows existing tools and scripts that expect those APIs to operate offline without modification.