LOGOS components~/architectureBrowse the work

Type a section name and press Enter. Use “help” for the list.

Figures: softened for dark viewingDisplay only; image files are unchanged.

LOGOS: components and mechanisms

The system’s component roles remain useful as the proposed cognitive core changes.

LOGOS is the larger system behind much of this work. Its earlier cognitive mechanism centered on a knowledge graph. I’m now investigating vector-symbolic representations through Cliste and related experiments, while retaining the larger architectural questions.

Cliste is intended to become Sophia; her earlier architecture and interfaces can change around the findings. Existing components may remain useful where they fit. Embodiment provides a way to test expectations through action. Acquired representations could let perception and memory guide those actions, with language services contributing where they help.

Sophia

The cognitive core: memory, learning, disposition, and agency. The aim is a continuing individual whose experience changes later perception and action.

Talos

Embodiment: the connection to perception and action. A central part of the intended system, beyond the current visual-memory experiments.

Hermes

Language and related services. The aim is to make less of the cognitive core depend on external language processing as the representation work develops.

Apollo

The interface for interacting with the system and inspecting what it is doing.

Cliste’s VSA experiments currently run separately from Sophia. LOGOS: persona, memory, and experiments →

Earlier graph-based design and implementation

The earlier architecture combined implemented services with proposals about how they could support cognition. The diagram records their intended connections; it does not establish an adaptive cognitive system.

System Overview

The graph-based design connected entities, relations, types, and persona records in the HCG. Sophia's diary retains beliefs, decisions, observations, and reflections; Apollo can retrieve recent entries for a later conversation. Language models contribute extraction and interpretation, while the broader research asks how experience could shape retained knowledge, perception, and disposition.

Memory across timescales

The design distinguishes ephemeral state and mood, short-term experience available for reassessment, and lasting knowledge and disposition. In-process state and persistent persona records exist; the complete promotion, expiration, and forgetting cycle remains to be implemented.

One Graph

The HCG connects entities, relations, types, and records about the system itself. A shared state envelope carries provenance for abstract, grounded, and emotional content. These are kinds of content, separate from the question of how long a memory is retained.

Revisable types

A research goal was to form candidate types from recurring structure, then revise membership, merge redundant types, or propose new ones. Whether a candidate type is useful requires evidence beyond clustering or a plausible name.

State changes perception

Separate experiments apply an internal-state matrix to embeddings and update it from new inputs. The hypothesis is that passing mood and lasting disposition change how an input is perceived. Connecting that transform to the persona diary and useful changes in continuing behavior remains open.

Earlier component diagram

TTS/STTchatNLP servicesannotation, knowledgeplanssensor datamotor datacommandsactuatorssensorsCWM embeddingsHCG queriesgraph staterequestsembeddingsUserApolloInterfaceHermesLanguage ServicesSophiaCognitive CoreTalosEmbodimentEmbodied AgentNeo4j + MilvusGraph + Vector Storage (Logos Infrastructure)

Core Components

Four services plus shared infrastructure, each with a distinct responsibility.

Sophia

Cognitive Core

The intended home of memory, learning, disposition, and decisions. Implemented pieces include graph maintenance, a persistent persona diary, and state records with provenance. Separate affect experiments investigate how internal state changes perception.

Apollo

Interface

Command-line and web interfaces for interacting with the services and inspecting state. The interface is a way to observe and direct the system, rather than evidence of its cognitive capabilities.

Hermes

Language Services

Language-service endpoints and adapters, including text embeddings and access to external language models. Speech support depends on the configured providers.

Talos

Embodiment

Sensor and actuator interfaces. The current Sophia inhabitant sends scripted motion commands through this connection; learning to act from observed consequences remains an open direction.

Logos

Infrastructure

Shared definitions, schemas, SDKs, and development infrastructure. Database configuration supports Neo4j graph storage and Milvus vector storage.

Technology Stack

Language & Runtime

The inspected service configurations target Python 3.12, with Poetry package configuration and Docker development environments. The Apollo web interface uses React and TypeScript.

Data Layer

The graph implementation uses Neo4j, with Milvus for vector storage. RDFLib and PySHACL support ontology representation and schema checks.

APIs & Communication

API code uses FastAPI and Pydantic. Apollo includes WebSocket connections, alongside a command-line interface built with Click and Rich.

Quality & CI

Repository configurations include pytest, coverage checks, Ruff, Black, and GitHub Actions. The scientific claims depend on the individual experiments and their controls.