Background Real-time surgical computer vision artificial intelligence (CV-AI) systems frequently perform well in development datasets; however, in-theatre deployment introduces safety, reliability, and workflow constraints that determine whether innovation can be integrated into routine operative practice. This design and integration pilot evaluated feasibility and operational stability of a multi-task CV-AI platform for laparoscopic cholecystectomy (LC), and captured early usability signals. Methods Twenty consecutive LC procedures were prospectively recorded using a standalone CV-AI workstation integrated into standard laparoscopic video infrastructure. Outputs were concealed from the operating team (surgeon-blinded). Primary feasibility endpoints were recording completeness, continuous runtime stability, and device-related safety events. Role-specific usability/acceptability ratings were assessed after the pilot by an in-theatre observer and two consultant surgeons. Results Analysable recording completeness was 90.0% (95% CI 68.3-98.8). Among analysable recordings (n = 18), reporting coverage was effectively complete (median 100.0% IQR 99.9-100.0). Two unplanned workstation shutdowns (10.0%) required manual restart (0.98 events per 10 operating hours), attributed to remediable thermal shutdown; estimated lost runtime was 3.8-6.7 minutes per event. No device-related safety events occurred (0%; 95% CI 0.0-16.8). Usability ratings suggested strong workflow fit and interface clarity; consultants perceived higher immediate value for documentation support than intraoperative decision-making, with high willingness to progress to surgeon-visible evaluation. Conclusions Standalone, surgeon-blinded in-theatre deployment of multi-task CV-AI during LC was feasible and workflow safe, with remediable reliability failures. These data support subsequent progression to clinician-facing evaluation and prospective validation.