Mika: In a Nutshell: App Manipulation Instead of Apps, an On-Device AI Agent for Android
wangzan101/Mika
About the project
Mika is an on-device AI agent that runs on Android devices. When a user enters a single sentence, Mika performs the necessary tasks within the device without requiring a separate server. App manipulation, web surfing, and script execution are all handled in the local environment, and data is sent only to the model provider configured by the user.
The core feature is its method of performing tasks on a 'hidden background screen.' Without interfering with the screen the user is currently using, Mika opens apps, clicks buttons, and inputs text in the background. It uses accessibility services and OCR to recognize the UI, reading and manipulating screens like a human, even in complex apps or web pages. In situations requiring direct intervention, such as logging in or making payments, it sends a notification to the user to hand over the task.
It supports a built-in browser and a Linux sandbox, enabling web page manipulation and the execution of Python and Node.js scripts. It embeds prior knowledge about specific apps or websites (entry points, general behavior) to improve task efficiency. It supports various LLM APIs, including DeepSeek, OpenAI, and Anthropic, and can also connect to local model servers like Ollama or LM Studio.
It can be installed on Android 11 or higher, but Android 13 or higher is required to use the app manipulation feature. This project, released under the Apache-2.0 license, is in its early beta stage, and its features and interface are changing rapidly. It is suitable for developers who want to delegate mobile automation tasks or repetitive app manipulation to AI.
wangzan101/Mika
An on-device AI agent for Android that operates apps and the web to get things done for you.
Kotlin
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.