
Kiara is an Android AI assistant that operates real mobile apps on behalf of users. It reads the Android accessibility tree, reasons about what is visible on screen, and performs actions such as tapping, typing, opening apps, sending messages, starting navigation, and scheduling tasks. Kiara runs in two modes: local mode keeps screen context private on-device, while cloud mode handles more complex reasoning. It supports multi-step goals like sending emails, messaging on WhatsApp or Instagram, navigating to destinations, booking transit, and setting reminders. A compact UI tree parser filters raw accessibility data into a smaller, model-friendly view with labels, roles, resource IDs, and interaction hints, letting smaller models pick correct actions without hardcoded app logic. User profiles let commands like "message Sarah" resolve to a saved contact, Instagram handle, or email without repeating details. Event-based rules such as "if Marco texts me, reply I'm busy" trigger the agent from matching notifications. Timed tasks run through a foreground service at the requested time. Kiara's goal is a privacy-aware, general-purpose Android agent that works with existing apps as they are, adapts to different interfaces, and completes real tasks with minimal friction. My aim is to scale it to enterprise level , companies can self-host models , this way they can automate repetitive tasks like PEC renewal.
19 May 2026