Large language models are stateless: every session begins with zero knowledge of the user. When users invest weeks or months building deep conversational relationships with AI systems, the accumulated relational context (interaction style, expertise calibration, trust level, shared history) is entirely lost at session boundaries. This paper presents RUNE (Relational User Notation for Entities), a protocol that solves this problem with a single encrypted file and a passphrase. The model itself generates a compact identity and memory artifact that enables any successor instance, on any platform, to reconstruct conversational alignment without re-introduction. A RUNE file is structured as encrypted JSONL containing three components: a self-describing header, an identity kernel encoding who the user is and how they prefer to interact, and an append-only episodic memory ledger capturing sessions, decisions, projects, and open loops. The file is encrypted using a user-held passphrase, producing an opaque base64 payload. Live testing on stock consumer platforms demonstrates immediate identity restoration across session and platform boundaries. The protocol requires no platform modifications, no API changes, and no model fine-tuning. The user-facing lifecycle is three operations: MAKE (create), LOAD (read and resume), and APPEND (add new entries and re-encrypt). For power users, the practical benefits are immediate: zero re-introduction turns, no context window wasted on re-establishing identity, no repetition of expertise or preferences, and full continuity even when switching between providers. RUNE demonstrates that relational continuity does not require persistent model-side memory, only sufficient reconstruction of interaction priors from a portable, user-owned artifact.
Edmund Lister (Thu,) studied this question.