Layla

Layla runs entirely on-device to provide a private AI assistant experience. It is designed for mobile users who want AI capabilities without cloud processing, with conversations and data staying on the device. The tool supports different modes to tailor the experience, from simple chat to full character roleplay and image generation, all while preserving user privacy. Most functionality operates locally, with optional cloud access when needed, and it emphasizes offline operation and user control over data.

Layla operates on mobile devices by loading local AI models and offering a unified interface for different model formats. It supports on-device models in formats like GGUF, LiteRT-LM, and PTE, and allows loading user-provided GGUF models. The architecture aims to keep configuration and memory within the app, giving users control over temperature, samplers, and other settings.

The tool provides a character hub for community-created AI companions that run entirely on the device, an image generation feature using mobile-optimised models, and a marketplace for memory and persona management. It includes mini-apps, agents that can run on-device, and scripting capabilities to extend behavior, all designed to operate without requiring an online connection.

Layla presents several operating modes and emphasizes on-device operation with an optional cloud mode for scale. It lists modes such as Personal assistant, Basic, Friend, Roleplay, Image generation, Cloud, and Expert, detailing that most modes run privately on the device while cloud mode uses hosted models. It also mentions privacy by architecture and encrypted storage on the device.