Open-source mobile UI patterns for agents

We are exploring a set of AI-first interface patterns designed to make agents feel more natural, transparent, and controllable on mobile devices. Today’s chat interfaces often hide what the system is doing, leaving users unsure about what an agent is thinking, accessing, or executing. Our goal is to introduce interaction patterns that make those processes visible and understandable while still feeling native to mobile platforms.

Chowder treats observability as a core part of the mobile chat experience rather than something hidden behind loading states. As the agent works, the app surfaces inline “thinking steps” and tool activity directly in the conversation, allowing users to see progress in real time. These updates reflect actions like fetching data or running tasks, then fade as the final response begins streaming, creating a more transparent and responsive interaction model. This approach extends beyond the chat itself through Live Activities, which make longer-running tasks visible at the system level. By combining streaming responses with real-time status and native mobile patterns, Chowder explores how AI interfaces can feel more observable, alive, and integrated into the operating system rather than functioning as a black box. Try it Here on Github