γ(3,4) Semantic Spacetime Agent

This project builds a reasoning agent around Mark Burgess’ Semantic Spacetime (SST) and Promise Theory, aligned with recent work on memory-centric AI agents (SciBORG, Zep, ArcMemo, Memento).

Core Ideas

• Avoid the Ontology Trap
  Ontologies are brittle; biotech shows us to link knowledge spaces dynamically instead of forcing a universal taxonomy.
  → We adopt local graphs with contextual edges rather than a rigid schema.

• Nodes & Arrows in Knowledge Graphs
  SST distinguishes:

  • Nodes: events, things, concepts (γ(3,4) triad).
  • Edges: NEAR, LEADS-TO, CONTAINS, EXPRESSES.
    Intentionality scores ∈ [0,1] weight nodes by information content (entropy).

• Promise Theory & Physics
  Promises express local commitments between agents/nodes. Physics is modeled as networks of promises, showing that causality and state emerge from cooperative commitments rather than imposed laws.

• Rescuing Property Graphs with SST
  Traditional property graphs lack causality and context. SST enriches them with temporal, semantic, and intentional structure—making graphs into reasoning substrates, not just storage.

Alignment with Current Research

• SciBORG (State and Memory is All You Need):contentReference[oaicite:3]{index=3}
  Modular LLM agents with finite-state automata (FSA) memory. Demonstrates that structured state memory dramatically improves robustness and tool use.
  → Our axiom layer enforces similar schema-driven state guarantees.

• Zep & Graphiti (Temporal Knowledge Graphs):contentReference[oaicite:4]{index=4}
  Knowledge as episodic → semantic → community subgraphs with temporal edges.
  → Mirrors SST’s layering of events → things → concepts.

• ArcMemo / Memento
  Memory systems that reinforce edges if used often, and allow selective forgetting.
  → Our reinforce_edge() boosts intentionality scores with usage, simulating neural consolidation.

Why This Matters

• Moves beyond ontology-based knowledge graphs → toward dynamic, intentional knowledge spaces.
• Provides stateful, causal reasoning while remaining lightweight and adaptable.
• Bridges theory (Burgess’ SST + Promise Theory) and practice (SciBORG, Zep, ArcMemo).

Implementation

• Graph extraction (graph_agent) builds γ(3,4) graphs from text.
• Reasoning agent (question_agent) traverses graphs, prioritizing high-intent nodes.
• Axiom layer (validate_graph) ensures valid knowledge graphs:
  • No causal cycles (LEADS-TO must be acyclic).
  • No NEAR-only links between events.
  • No self-loops.
• Reinforcement updates intentionality over time (Memento-style).

See example02_agent.py for full implementation.

Visual Overview

graph TD
  A["Event: toxic takers reduce activity (0.9)"] -->|LEADS-TO| B["Event: market makers narrow spreads (0.9)"]
  B -->|EXPRESSES| C["Concept: confidence in liquidity (0.9)"]
  A -.NEAR.-> D["Thing: toxic takers (0.8)"]
  B -.CONTAINS.-> E["Thing: market makers (0.8)"]

Loading

Quick Start

# Install dependencies  
uv sync

uv run scripts/example02_agent.py