Emerging reconfigurable devices are fast gaining popularity in the search for next-generation computing hardware, while ferroelectric engineering of the doping state in semiconductor materials has the potential to offer alternatives to traditional von-Neumann architecture. In this work, we combine these concepts and demonstrate the suitability of reconfigurable ferroelectric field-effect transistors (Re-FeFETs) for designing nonvolatile reconfigurable logic-in-memory circuits with multifunctional capabilities. Modulation of the energy landscape within a homojunction of a 2D tungsten diselenide (WSe 2 ) layer is achieved by independently controlling two split-gate electrodes made of a ferroelectric 2D copper indium thiophosphate (CuInP 2 S 6 ) layer. Controlling the state encoded in the program gate enables switching between p, n, and ambipolar FeFET operating modes. The transistors exhibit on−off ratios exceeding 10 6 and hysteresis windows of up to 10 V width. The homojunction can change from Ohmic-like to diode behavior with a large rectification ratio of 10 4 . When programmed in the diode mode, the large built-in p−n junction electric field enables efficient separation of photogenerated carriers, making the device attractive for energy-harvesting applications. The implementation of the Re-FeFET for reconfigurable logic functions shows how a circuit can be reconfigured to emulate either polymorphic ferroelectric NAND/AND logic-in-memory or electronic XNOR logic with a long retention time exceeding 10 4 s. We also illustrate how a circuit design made of just two Re-FeFETs exhibits high logic expressivity with reconfigurability at runtime to implement several key nonvolatile 2-input logic functions. Moreover, the Re-FeFET circuit demonstrates high compactness, with an up to 80% reduction in transistor count compared to standard CMOS design. The 2D van de Waals Re-FeFET devices therefore exhibit promising potential for both More-than-Moore and beyond-Moore future of electronics, in particular for an energy-efficient implementation of in-memory computing and machine learning hardware, due to their multifunctionality and design compactness.