There have been recent advances in brain-computer interfaces for post-stroke rehabilitation. In particular, compact and embedded brain-computer interface systems with neuromuscular electrical stimulation have been developed by industry and academia, and some of them can potentially be used at the bedside. However, limited studies have demonstrated their safety and feasibility for treatment in subacute stroke patients. The aim of this pilot study was to first develop a brain-computer interface system for subacute stroke inpatients that is usable at the bedside and to show the safety and feasibility using a small cohort of inpatients. Four hemiplegic stroke inpatients in the very early phase (7-24 days from stroke onset) participated in this study. The portable brain-computer interface system shows the amplitude of sensorimotor rhythms extracted from scalp electroencephalograms in real time. Patients attempted to extend the wrist on their affected side, and neuromuscular electrical stimulation was applied only when the brain-computer interface system detected significant movement intention-related electroencephalogram changes. Each brain-computer interface training lasted 40 minutes. On average, 120-200 training trials of the wrist extension task were successfully and safely conducted over 3.3 days (range 2-4 days) with the bedside brain-computer interface system. Furthermore, electroencephalogram results showed a new significant event-related desynchronization in the damaged hemisphere after training. These results clearly show the proposed bedside brain-computer interface system's safety and feasibility and also demonstrated electrophysiological plasticity in the damaged hemisphere in subacute patients with post-stroke hemiplegia. Larger clinical studies are needed to identify the brain-computer interface system's clinical efficacy and its effect size in the subacute post-stroke patient population.