MindForm — Motivation and Drives: Basic Psychological Needs as a Fast Motivational State in LLM Agents

Personality architectures for language-model agents typically represent dispositional structure — traits, values, characteristic construal — but not momentary motivational pressure, and are consequently able to describe what an agent is like without explaining why it acted now. This paper documents the Motivation and Drives node of the MindForm architecture, which supplies that missing quantity. Following Self-Determination Theory, the node represents three basic psychological needs — autonomy, competence and relatedness — each maintained as a tension in the unit interval. Tension falls under events that feed a need and rises, at higher gain, under events that frustrate one, implementing SDT's asymmetry between need satisfaction and need frustration; between events it relaxes toward a resting weight projected from the character's formed Schwartz values, so that which needs matter is itself earned through experience. Two structurally distinct projections separate what an event teaches a character to value from what it nourishes. The node operates entirely without a language model, a principled consequence of needs carrying no propositional content, and additionally grounds a motivated-retrieval mechanism in which active needs re-rank recalled memory multiplicatively over semantic similarity. With this node the architecture closes its first complete feedback cycles: formed values set need weights, active needs tilt appraisal and thereby regulate how strongly any experience forms the character, and the tilted appraisal is the input from which values continue to form. We claim the existence of these cycles, which are verifiable in source, and not their stability. All quantities within them are bounded, but we specify a concrete structural path along which sustained need frustration could produce monotone drift in the value profile with no variable leaving its range, and identify the longitudinal experiment required to measure it.