Satellite detection of thermal anomalies or active fires is conventionally carried out by exploiting the thermal signal around 3.8 mm wavelength window in the shortwave-infrared (SWIR) part of the spectrum, and its contrast to the infrared measurements around 11.2 mm, produced by fires. The operational and near-real time active fire detection products are being produced globally on a daily basis from NASA’s MODIS (1-km) and NOAA-NASA’s Suomi-NPP/VIIRS sensors (375 meters), providing once or twice a day global view of fire system since 2000 and 2012, respectively. The active fire detection is continued with the application of the similar algorithm to the VIIRS sensor onboard NOAA-20 and NOAA-21 platforms. The hourly active fires and hot smoke detection dataset for the Punjab and Haryana states of India for the post-monsoon months of October and November presented here is derived using the observations made by the Advanced Meteorological Imager-AMI aboard South Korean GEOKOMPSAT-2A satellite on the geostationary orbit. A preliminary algorithm that utilizes the spectral contrast in the brightness temperature between the 3.8 mm and 11.2 mm wavelength windows was applied to GK2A-AMI observations to deduce hourly fires and hot smoke pixel detection for measurements made between 12:00 to 18:00 Indian Standard Time (IST). The thermal measurements made between 9:00 to 12:00 IST hours are used to normalize the observations made during the rest of the day and up to 18:00 IST hour to account for the day-to-day variabilities in the brightness temperatures. The GK2-AMI fire detection dataset is provided at its native spatial resolution, which is ~2 km at sub-satellite point, but increases to ~ 3.8-4.0 km over the northwestern India due to slant viewing geometry. This dataset is associated with the following publication: Jethva, H. Timing the Flames: Geostationary Satellite Detection of Diurnally Shifting Stubble Burning in Northwestern India, Remote Sens. 2026. 18(10), 1506. https://doi.org/10.3390/rs18101506.