Barcelona-based quantum developer Qilimanjaro Quantum Tech has released QiliSDK 0.2.0, an open-source Python framework designed to bridge the structural divide between digital gate-based programming and analog Hamiltonian time evolution. Most quantum software development kits restrict developers to a single physical computing paradigm—either compiling standard quantum circuit logic or managing continuous analogue energy schedules. QiliSDK 0.2.0 establishes a single unified, backend-agnostic API. This abstraction tier enables researchers to write a single high-level codebase and toggle execution between local CPUs, accelerated GPUs, or real digital and analog Quantum Processing Units (QPUs) by modifying a single line of configuration code. Three-Tier Architecture and Native Multi-Paradigm Abstractions The framework is organized into three distinct operational layers: Primitives, Functionals, and Backends. The Primitives layer provides a foundational toolkit containing pre-built variational ansatz blocks for digital circuits, alongside continuous time-evolution schedule modules for analog systems. These modules feed into a core quantum-tensor type—now optimized as a native C++ module named QTensor—that handles high-speed state preparations, observables, and partial traces. The Functionals layer uses a standardized backend.execute(functional) method to normalize diverse routines like variational loops or analog annealing runs. This unified interface interfaces directly with the Backends layer, which maps computational tasks across classical emulators like QuTiP, NVIDIA graphics processors, or Qilimanjaro’s cloud-linked and on-premise physical quantum computers. GPU Acceleration via NVIDIA CUDA-Q Integration To support large-scale quantum emulation before dispatching code to physical QPUs, Qilimanjaro has integrated NVIDIA CUDA-Q directly into the framework via the CudaBackend class. The upgrade allows users to leverage the parallel processing power of graphics cards to track expanding quantum states, which grow exponentially with qubit count and quickly exhaust standard CPU memory limits above 25 qubits. The CUDA-Q wrapper automatically configures multi-GPU pooled execution and handles advanced multi-node tensor-network contractions. For analog operations, the backend translates time-dependent schedules into optimized operators,