This paper proposes the Raised Learning Companion Agent (RLCA), a conceptual and design-science framework for AI-era education. The central problem addressed is not whether AI can explain, tutor, or generate feedback, but whether students can repeatedly return to AI as a guided learning practice rather than use it episodically as an answer machine. The framework identifies six structural lock-ins of contemporary education: memorization-centered learning, rank-based comparison, credential-first evaluation, lecture-only teaching, AI prohibition, and digital inequality. RLCA responds by translating digital-pet and companion-interface mechanisms—persistent identity, care, visible growth, ownership, and return motivation—into a teacher-mediated educational architecture focused on questioning, verification, expression, reflection, and offline learning. An RLCA is defined by five minimum criteria: persistent companion identity, student-shaped learning growth, questioning and verification support, teacher-mediated governance, and offline learning re-anchoring. The paper explicitly rejects teacher replacement, screen-time maximization, and unconstrained emotional companionship. Retention is treated only as a mediator of repeated learning opportunity, not as proof of learning. The paper does not report completed classroom validation. Instead, it defines the RLCA design category, safety boundaries, system architecture, governance requirements, and a staged evaluation protocol. Future pilots should compare tool-based AI, AI tutors, full RLCAs, and ablation conditions across question quality, verification behavior, expression, secondary academic outcomes, teacher trust, equity, privacy comprehension, and dependency risk. In short, the paper reframes AI-era education from “ban AI or use AI” to “raise safe learning companions that help students become better questioners, verifiers, and self-directed learners.”