Android agent Mika gains 7 495 stars in 9 days and runs on a hidden screen

Mobile agents still have a basic usability problem: while AI taps away at the screen, you cannot use your phone. It takes over the display, leaving you to wait or get in its way. Mika, a project by wangzan101, tackles this head-on: the agent works on a hidden second screen, leaving the main screen to you.

A post by @ReadmeRadar suggests enormous interest in the idea: 7 495 GitHub stars in nine days. That signals demand rather than quality, but the demand is telling.

What we know about Mika

According to its description, Mika launches and controls your apps on a hidden second screen while you continue using your phone. Its code is available under Apache-2.0, meaning you can fork it, integrate it and use it in commercial products without unpleasant surprises.

The second important detail is the architecture. Mika has no server of its own. You supply a model API key or sign in through ChatGPT, xAI or OpenRouter. For users, this means requests and screenshots do not pass through a separate backend run by the project, and you pay the model provider directly. For developers, it means less reliance on a third party and the freedom to choose a model that fits the task and budget.

A simple example

Imagine asking the agent to order groceries in a delivery app and compare prices at two stores. Previously, your phone screen would become someone else's workspace for several minutes: trying to tap anything yourself was pointless. With a hidden screen, you can reply to messages while the agent browses the catalog on a second, invisible display.

The post does not explain exactly how this second screen is implemented. The wireless debugging requirement suggests that control uses ADB-level mechanisms, but this is our inference, not a confirmed fact. Before building anything substantial on it, inspect the source code and verify how it works.

Limitations

The post's author points out the catch: controlling apps requires Android 12+ with wireless debugging enabled. That is a significant barrier for an ordinary user. Network debugging is a developer mode, and enabling it expands the attack surface. An agent that sees your app screens and presses buttons could potentially reach banking apps, email and messaging services.

Then there are the usual agent problems: the model may misread an interface, tap the wrong place or get stuck on a CAPTCHA. Some apps block automation or hidden screens altogether. The post says nothing about Mika's reliability, and stars do not measure it.

What to do with this

• Mobile agent developers. Examine how the project handles the background screen and API keys. The license allows you to reuse its solutions.
• Product managers. An agent that leaves the user's screen available offers a clear competitive advantage for any phone assistant. Consider including this in your requirements.
• Founders. The model of having no server of your own and asking users to supply their keys reduces your costs and legal risks, but complicates onboarding. Wireless debugging will deter mainstream users.
• Anyone trying it out. Run it on a separate device or a test account, and keep the agent away from financial apps until you have verified exactly what it does.

Mika is not a proven product, but it is a useful indicator of where things are heading. The market wants mobile agents that work alongside people instead of taking over their screens. Treat the project as a reference and an opportunity to experiment, rather than a production-ready solution.