This paper presents the methodology and outcomes of a comprehensive dataset collection using ESP32-Nodemcu devices and the ESP32-CSI Toolkit. The dataset, designed to explore the capabilities of Channel State Information (CSI) in distinguishing human activities, was collected in a controlled indoor environment under three scenarios: single-user, two-user, and three-user setups. The experimental setup involved 80+ participants performing six carefully selected activities, ranging from subtle hand movements to dynamic full-body actions, ensuring diverse motion patterns and environmental interactions. The data acquisition process employed a transmitter-receiver configuration to capture fine-grained variations in CSI caused by human motion. By prioritizing distinct activities and managing variability, this dataset provides a robust foundation for develop- ing and validating multi-human activity recognition models. The work aims to advance the understanding of non-intrusive, device- free systems, offering valuable insights into the potential of WiFi signals for human activity recognition in complex environments.