Current digital financial advice often faces a trade-off between the rigid personalization of robo-advisors and the generic nature of ungrounded conversational agents. Consequently, existing solutions struggle to build the deep, competence-based trust necessary for broader user adoption. To address this socio-technical gap, this paper follows a Design Science Research (DSR) methodology according to Peffers et al. to design, build, and evaluate a hyper-personalized financial co-pilot. The web-based artifact leverages a stateful, session-based memory to ground a Large Language Model’s responses, enabling a deep, conversational understanding of a user’s unique goals. We evaluated the artifact in a qualitative focus group to assess its impact on trust and perceived personalization. The evaluation reveals that while stateful memory significantly enhances perceived competence, it introduces critical challenges regarding robustness and the depth of inquiry required to satisfy user expectations of hyper-personalization. The study contributes a set of refined Design Features (DFs) that extend beyond technical features to address the socio-technical requirements of institutional transparency and graceful failure management for the design of trustworthy financial co-pilots.