Open-Source Mika Controls Android Apps Invisibly in the Background
Mika is an open-source Android AI agent that operates apps on a hidden background screen while you use your phone Over 2,000 GitHub stars within days, Apache-2.0 licensed, written in Kotlin and available on GitHub Bring-your-own-key design supports OpenAI-compatible, Anthropic and Gemini endpoints
Takeaways
- Mika is an open-source Android AI agent that operates apps on a hidden background screen while you use your phone
- Over 2,000 GitHub stars within days, Apache-2.0 licensed, written in Kotlin and available on GitHub
- Bring-your-own-key design supports OpenAI-compatible, Anthropic and Gemini endpoints with no Mika server in between
- Includes a built-in desktop-class browser and per-conversation Linux sandbox for Python, Node and shell
- Ask vs Autonomous modes gate money-spending actions; logins and captchas always handed back to user
- Requires Android 13+ for tasks, arm64 for sandbox, UI currently Chinese-only, tested mainly on Android 14
Mika is an open-source Android agent that operates existing apps through taps, swipes, and text input. Its automation runs on a separate background display, leaving the foreground available for normal phone use. The Apache-2.0 project is written in Kotlin with Jetpack Compose and has 26 GitHub stars.
How the background display works
Mika launches target apps on a separate display and reads their interfaces while a small live preview appears in the conversation view. This design lets it operate apps such as WeChat, Xiaohongshu, BOSS Zhipin, and Bilibili without taking over the screen currently in use.
Its perception stack starts with Android accessibility services, which expose a structured tree of buttons, text fields, and other controls. Bundled optical character recognition handles text inside custom interfaces that omit useful accessibility data. An optional vision model can inspect screenshots to identify controls that neither channel exposes.
The interaction loop resembles desktop computer-use agents from Anthropic and OpenAI: a model observes the current state, chooses an action, checks the result, and repeats. Mika applies that loop to consumer Android apps on a physical device.
Local execution, external inference
Model requests go directly from the device to the provider configured by the user. Mika operates no intermediary backend and supports presets for DeepSeek, Qwen, Kimi, Zhipu, OpenRouter, Anthropic, and Gemini, along with compatible OpenAI-style endpoints.
Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.
Read original report
