This article proposes a solution to improve office chairs (referred to as IoT chairs) based on IoT technology and LSTM (Long Short – Term Memory) neural networks to monitor and promptly warn via the Internet about questions of abnormal health status of office staff. An IoT circuit with the MCU-ESP8266 module is used to collect weight and an accelerometer sensor embedded in the chair, which can communicate with a computer to monitor the searing time of the user and warn by sound for prolonged sitting. LSTM neural networks built on MATLAB is trained by deep learning techniques to track inappropriate postures of people sitting in chairs, through analyzing signals from sensors. Experiment results on many different scenarios show that the accuracy of capacity of reminding about the status of prolonged sitting is 100% and reliability of the capacity of detecting and warning abnormal health conditions is 94%. Experiments also show that the ability to complete IoT chairs for a popular application is completely feasible.