Overcoming the App Shortage
One of the main challenges for new operating systems remains porting existing software: rewriting millions of lines of code natively from Android or iOS is too expensive. The situation was further complicated by a common argument: modern AI models know almost nothing about Aurora OS because they were trained on foreign ecosystems.
To test whether artificial intelligence could handle this task, the creators of Aurora OS conducted an open experiment together with independent mobile engineer Alexey Gladkov.
Selection of Target and Technology Stack
The popular open-source enterprise messenger Mattermost was chosen for the test. The source project contained over 220,000 lines of code in React Native, Java, and Kotlin, along with numerous plugins and a non-standard architecture.
Flutter, which has official support on Aurora OS, was selected as the target framework. This allowed the team to supply the Claude AI agent with up-to-date documentation and context with integration examples.
Tools and the Porting Process
The migration was carried out using specialized tools:
- Flutter Aurora SDK for macOS — enabled running and building the project on developer workstations.
- claude-in-mobile — a utility allowing the neural network to interact with a real device and an emulator, verifying UI screen readiness rather than just the source code.
- Custom agent — a script loaded with the platform knowledge base that managed the migration process.
The migration process was cyclical: the agent drafted a work plan, gradually recreated screens and logic in Flutter, and then compared the interface with the Android original until complete visual and functional parity was achieved.
Results and Release in RuStore
The final phase included end-to-end testing: authenticating with the Mattermost server, sending messages, and executing user scenarios. As a result, the application retained its original visual style, ran stably on the test device, and successfully passed moderation in the RuStore catalog.
Conclusions and Platform Growth Areas
The experiment confirmed that the absence of the OS in global LLM datasets ceases to be an obstacle if the agent's context is enriched with documentation and debugging tools. Nevertheless, developers identified key areas for platform growth:
- The need to build an AI-ready interface — a standardized bridge (similar to Android Debug Bridge) for programmatic UI control by AI.
- Adapting documentation to machine-readable formats to simplify the work of AI agents.
- Further expanding the repository of ready-made Flutter plugins for Aurora OS.
This successful case demonstrates that neural network tools can significantly lower the barrier and cost of porting enterprise software to the domestic mobile OS.