This paper presents a noninvasive neuromotor interface for human-computer interaction, leveraging surface electromyography (sEMG) to enable computer input without the need for intermediary devices. The authors developed a highly sensitive and robust hardware platform that can be easily donned and doffed, sensing myoelectric activity at the wrist and transforming it into computer input. They paired this device with an infrastructure optimized for collecting training data from thousands of consenting participants, allowing the development of generic sEMG neural network decoding models that work across many people without the need for per-person calibration. The models demonstrated high performance in various tasks, including continuous navigation, discrete gesture detection, and handwriting transcription, achieving median target acquisitions per second of 0.5, 0.9, and 17.0 adjusted words per minute, respectively. The study also explored the scalability and generalization of these models, showing that performance improves with larger training datasets and that personalizing models to individual participants can further enhance performance. This work represents a significant advancement in the field of neuromotor interfaces, providing a versatile and accessible solution for human-computer interaction.This paper presents a noninvasive neuromotor interface for human-computer interaction, leveraging surface electromyography (sEMG) to enable computer input without the need for intermediary devices. The authors developed a highly sensitive and robust hardware platform that can be easily donned and doffed, sensing myoelectric activity at the wrist and transforming it into computer input. They paired this device with an infrastructure optimized for collecting training data from thousands of consenting participants, allowing the development of generic sEMG neural network decoding models that work across many people without the need for per-person calibration. The models demonstrated high performance in various tasks, including continuous navigation, discrete gesture detection, and handwriting transcription, achieving median target acquisitions per second of 0.5, 0.9, and 17.0 adjusted words per minute, respectively. The study also explored the scalability and generalization of these models, showing that performance improves with larger training datasets and that personalizing models to individual participants can further enhance performance. This work represents a significant advancement in the field of neuromotor interfaces, providing a versatile and accessible solution for human-computer interaction.