We were working with AI models before coding agents became part of every team's toolkit. In 2024 we started running models directly on phones with React Native ExecuTorch, our open-source library for local inference. Today, our AI models are downloaded more than 600,000 times a month on Hugging Face.
When coding agents arrived, we ran into their limits on our own projects. An agent could write a screen in minutes and still had no way to open the app and see whether it worked. So we built Argent, an open-source toolkit that lets agents run, debug and profile iOS and Android apps on their own. Our engineers work with Claude Code and Codex every day, across Swift, Kotlin, React Native, TypeScript, Elixir and Python codebases.