This work introduces a novel circuit-level implementation of a fully memristor-based Liquid Spiking Neural Network (LSNN), connecting liquid neural computation, spiking neuromorphic systems, and emerging memristive hardware. The architecture adapts the concept of Liquid Neural Networks (LNNs), originally explained by Ramin Hasani et al. in [1] and [2], to the spiking domain, where temporal adaptability arises directly from device and circuit dynamics rather than from fixed parameters. The proposed network is more “liquid” than the original LNN model, as both neuron units and synaptic behaviors are input- dependent and realized with memristors. At the neuron unit level, a liquid Leaky Integrate-and-Fire (LIF) model is introduced, in which the effective membrane time constant is controlled by a memristive element rather than a standard resistor-capacitor pair like an original LIF model has [3]. This allows for continuous adaptation of the integration dynamics based on spiking activity, enabling neurons to function across multiple temporal scales. The corresponding circuit implementation is shown in Fig. 2. At the synaptic level, traditional static weights are replaced with memristive conductances arranged in crossbar arrays (Fig. 1). These synapses are trained using a supervised learning method (not STDP), in which conductance values are set by an offline-trained model. The LSNN uses a fully recurrent topology (Fig. 1) and burst-based spike encoding. To maintain temporal information and stabilize recurrent computation, spike-pattern flip-flop circuits are introduced (Fig. 2), which allow reconstruction and reinjection of past spiking states. The system is specifically tested for sequence prediction tasks, as shown in Fig. 3, where repeating input patterns yield consistent, temporally structured spike outputs. The LSNN is implemented and validated through circuit-level simulations using the IHP MEMRES-SG13S design kit. Results show accurate sequence prediction, robustness to device variability, and energy efficiency, emphasizing the potential of memristor-based LSNNs for adaptive, real-world neuromorphic computing applications. [1] R. Hasani, M. Lechner, A. Amini, D. Rus, and R. Grosu, Liquid time-constant networks, Dec. 14, 2020. [2] R. Hasani et al.,“Closed-form continuous-time neural networks,” Nature Machine Intelligence, vol. 4, no. 11, pp. 992–1003, Nov. 2022. [3] A. Ben Abdallah and K. N. Dang,“Neuromorphic system design fundamentals,” in Neuromorphic