What does this research mean for the field?

The OLYMPUS architecture provides a structured, locally-executable pipeline of specialized systems and psychologically-modelled memory layers to enable AI companions with persistent emotional states and coherent identities. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The aim is to present a structured architecture for AI companions that can manage emotions and memories effectively.

May 17, 20260 citationsOpen Access

View Full Paper

OLYMPUS: Companion Intelligence Architecture

JBJesus Antonio Flores Bernal

Key Points

The aim is to present a structured architecture for AI companions that can manage emotions and memories effectively.
Describes a framework with eleven systems, each with specific functions, including emotional management and memory persistence.
Implements a continuous emotional intensity scale from 1.0 to 4.0 and utilizes four psychological layers for memory storage.
Runs locally on consumer hardware, ensuring no dependency on cloud services.
Establishes a comprehensive architecture for AI companions with defined responsibilities for each component.
Demonstrates the capability of maintaining coherent character identity and emotional states across sessions.
Provides a reference for developers to build effective AI companion systems.

Abstract

OLYMPUS is a companion intelligence architecture — a structured pipeline for building AI companion systems that maintain persistent emotional state, layered relationship memory, and coherent character identity across sessions.The framework comprises eleven named systems (PSYCHE, SOPHIA, HERMES, MOIRA, POTHOS, ANANKE, LOGOS, KRISIS, MNEME, ARCHIVIST, SISYPHUS) each with a single defined responsibility, plus an external behavioral auditor (THEMIS). Emotional intensity is managed as a continuous float (1.0–4.0) with structural form mandates that scale with intensity. Memory persists across sessions through four psychologically-modelled layers (SUPEREGO, EGO, ID, THYMOS).The reference implementation runs fully locally on consumer hardware with no cloud dependency. This document is the framework specification. Source code is not included.Keywords: AI companion, state machine, persistent memory, emotional escalation, character architecture, local LLM, companion design

Demander à l'IA

Bookmark

View Full Paper

Demander à l'IA

Bookmark

View Full Paper

OLYMPUS: Companion Intelligence Architecture

Key Points

Abstract

Cite This Study