Providing thermal comfort in office buildings is crucial for maintaining productivity and wellbeing, especially in free-running offices prone to overheating during heatwaves. Accurate temperature forecasting is essential for predicting and mitigating such conditions. Building on prior research on time series forecasting for indoor temperature prediction, this study extends these methods by deploying linear AutoRegressive with eXogenous inputs (ARX) models on edge devices for real-time, on-site forecasts. This paper introduces a system capable of delivering precise 24-h indoor temperature forecasts using cost-effective edge devices for widespread adoption. The forecasting model integrates real-time sensor data with historical and future meteorological forecasts from the MeteoBlue service, and the model executes directly on the edge device. By using a Raspberry Pi as the computational unit, the study showcases the feasibility of processing high-frequency sensor data and generating accurate predictions locally without relying on cloud-based solutions. The model was validated during a heatwave in Graz, Austria, in the summer of 2024, achieving a mean absolute error (MAE) of 0.681 °C over a 24-h forecast horizon. Additionally, the forward-looking forecasts were applied to predict thermal comfort using adaptive criteria from the EN 16798-1 standard, enabling proactive management of occupant well-being in response to dynamic indoor conditions. This study highlights the potential of edge-based temperature forecasting to transform building management systems, offering real-time, on-site decision-making to maintain thermal comfort while addressing the challenges of climate change and increasing heatwaves.