Ontology Module - Clean-Room Knowledge Graph

Zero external dependencies. SQLite-based. Sovereign Stack compatible.

Status: ✅ INTEGRATED

Completed: 2026-04-27
Lines of code: ~1,100 (schema + engine + NL interface)
Test coverage: 7 query patterns verified

Files

Quick Start

cd /home/hoffmann_admin/.openclaw/shared/ontology

# Test the engine
python3 -c "from engine import OntologyEngine; e = OntologyEngine(); print(e.get_stats())"

# Test natural language queries
python3 -c "from natural_language import ask_ontology; print(ask_ontology('who works on icarus'))"

# CLI entry points
python3 -m ontology.engine stats
python3 -m ontology.engine query "what projects is socrates working on"
python3 -m ontology.engine export ~/Obsidian/Ontology

Verified Query Patterns

Bootstrap Data (Pre-loaded)

Board:
- person:matt — Director
- agent:wadsworth — Chief of Staff (glm-5.1:cloud)
- agent:socrates — Backend Architect (kimi-k2.5:cloud)
- agent:daedalus — UX Architect (deepseek-v4:cloud)

Family:
- person:aundrea — Matt's spouse
- person:sully — Son, 1st grade
- person:harper — Daughter, pre-k

Projects:
- project:icarus — Phase 5, family assistant
- project:costco_route — Stable, shopping optimizer
- project:ontology — Building, knowledge graph

Relations:
- serves, works_on, coordinates, child_of, spouse_of

Python API

from ontology.engine import OntologyEngine, get_engine
from ontology.natural_language import ask_ontology

# Direct CRUD
engine = OntologyEngine()
engine.add_entity("concept:ghost_bug", "concept", "Ghost Bug", {"severity": "critical"})
engine.add_relation("project:icarus", "concept:ghost_bug", "affected_by")

# Natural language
result = ask_ontology("what projects is wadsworth working on")
# Returns: "📋 Projects involving wadsworth:
#   • Icarus (phase_5)
#   • Ontology (building)"

# Export to Obsidian
engine.export_to_obsidian(Path("~/Obsidian/Ontology"))

Schema

Entities

• id (PK): type:name format (e.g., agent:wadsworth)
• type: person, agent, project, event, document, skill, system, location, organization, concept
• name: Human-readable label
• properties: JSON flexible attributes
• source: Where we learned this (manual, system, inferred)
• confidence: 0.0-1.0 certainty
• created_at, updated_at: Timestamps

Relations

• from_id, to_id, relation_type (composite PK)
• properties: JSON relation attributes
• confidence: 0.0-1.0
• created_at: Timestamp

Observations

• Temporal facts about entities
• entity_id, observation (text), observed_at, source, confidence

Obsidian Export

Generates Markdown files with wikilinks:

# Wadsworth
**Type:** agent

## Properties
- role: chief_of_staff
- model: glm-5.1:cloud

## Relations
- → [[person:matt|Matt]] (serves)
- → [[project:icarus|Icarus]] (coordinates)

Sovereign Stack Principles

• ✅ No cloud dependencies — SQLite only
• ✅ No new graph DB — Plain SQL, no Cypher/Gremlin
• ✅ Natural language — No query language to learn
• ✅ Human-in-the-loop — All modifications require explicit intent
• ✅ Exportable — Plain text, Obsidian-compatible
• ✅ Clean-room — Zero code from ClawHub skill

Integration Points

Wadsworth can use this for:
- Cross-referencing projects and agents
- Family member lookups (e.g., "who needs to know about Sully's field trip?")
- Contextual awareness (e.g., "Socrates worked on Icarus → use backend patterns")
- Memory lineage (observations trace back to source)

Future extensions:
- Link to daily logs (memory/YYYY-MM-DD.md)
- Auto-extract entities from conversations
- Confidence decay for stale observations
- Temporal queries ("what was Socrates working on last month?")

License

Same as OpenClaw — MIT or as specified by Director.

Clean-room implementation. 1,100 lines. Zero dependencies. Sovereign by design.