Architecture

A graph-based cognitive architecture for causal reasoning and autonomous action

System Overview

LOGOS represents knowledge as structured graphs rather than linguistic tokens, enabling formal validation and transparent reasoning about actions and consequences.

Non-Linguistic Cognition

The agent reasons through causal graphs, not token sequences. It understands why actions lead to outcomes, not just that they correlate in text.

Causal World Model

Knowledge is structured as entities, states, and processes with explicit causal relationships, the kind of understanding infants develop before language.

Evolving Self-Model

The agent maintains a persistent persona: a graph-based representation of its own beliefs, experiences, and reasoning that grows and changes over time.

Formal Verification

Plans are validated against ontological constraints before execution. There are no hallucinations because any required precondition must be met.

System Architecture

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 central orchestrator and reasoning engine. Manages the causal world model, planning, and decision-making. Sophia owns the hierarchical causal graph and coordinates all cognitive operations.

Apollo

Interface

User-facing layer providing command-line and web interfaces. Visualizes agent state, graph structure, and plan execution. Communicates with Sophia to send goals and display results.

Hermes

Language Services

Stateless language processing utilities. Provides speech-to-text, text-to-speech, text embeddings, and a gateway to external language models. All endpoints are stateless by design.

Talos

Embodiment

Hardware abstraction layer for sensors and actuators. Executes plans generated by Sophia. Currently provides simulated interfaces; designed for future physical hardware integration.

Logos

Infrastructure

Meta-repository containing specifications, ontology definitions, SHACL schemas, and shared development infrastructure. Hosts the Neo4j and Milvus clusters used by all components.

Technology Stack

Language & Runtime

Python 3.11+ across all components, with Poetry for dependency management and Docker for containerization. The Apollo web interface uses React with TypeScript.

Data Layer

The Hierarchical Causal Graph lives in Neo4j, with Milvus providing vector storage for semantic embeddings. RDFLib and PySHACL handle ontology representation and constraint validation.

APIs & Communication

Services expose REST endpoints via FastAPI, with Pydantic for request/response validation. Real-time updates flow through WebSockets. The CLI uses Click and Rich for terminal interaction.

Quality & CI

pytest with coverage requirements, Ruff for linting, Black for formatting, and GitHub Actions for continuous integration across all repositories.