We present the system architecture of an integrated Ultrasound-guided High Intensity Focused Ultrasound (USgHIFU) system for image-guided surgery and temperature tracking in vivo. The system is capable of operating with multiple frontends. Current implementation has a SonixRP for imaging and a custom designed dual mode ultrasound array (DMUA) system (32Tx/32Rx) for imaging/therapy. The highlights of the system include a fully-programmable, multiple data stream capable data processing engine, and an arbitrarily programmable high power array driver that is able to synthesize complex beam patterns in space and time. The data processing engine features a pipeline-style design that can be programmed on-the-fly by re-arranging the pre-verified GPU-accelerated high performance pipeline blocks, which cover an extensive range from basic functions such as filtering to specialized processing like speckle tracking. Furthermore, the pipeline design also has the option of bringing in MATLAB (Mathworks, Natick, MA, US) as part of the processing chain, thus vastly increase the capability of the system. By properly balancing the processing load between GPU-enabled routine and MATLAB script. This allows one to achieve a high degree of flexibility while meeting real-time constraints. Results are presented from in vivo rat experiment. Where low dose of therapeutic ultrasound was delivered into the hind limb of the Copenhagen rats using DMUA and temperature was tracked using a linear probe (HST, Ultrasonix). The data is processed in realtime with MATLAB in the loop to perform temperature regularization. Results show that we can reliably track the low temperature heating in the presence of motion artifacts (respiration and pulsation).