Automating trading decisions has been a pursuit of researchers and practitioners alike for decades. We contribute to the literature focusing on “pattern based” strategies. Dynamic time warping is used to group similar patterns into a representative category, while the method of continual learning augmentation is used to maintain the set of patterns used for decision-making. Thus, we implement a novel approach to pattern-based trading, utilising adaptive memory structures to enable adaptability of agent decision making and overall agent performance. Two new online pattern-based trading agents are introduced and tested on two-sets of historical cryptocurrency data, for the BTCUSDT pair over the periods of 2017–2023 and 2023–2024. We compare our newly formulated agents against an established baseline of rule-based agents, thereby comparing the relative profit generating abilities of a wide range of agents.