license: mit
language:
- en
tags: - knowledge-graph
- ai-agents
- constitutional-ai
- world-model
- agent-memory
- neo4j
- guardrails
- persistent-memory
- odei
pretty_name: ODEI Constitutional World Model
size_categories: - n<1K
ODEI Constitutional World Model
Production knowledge graph for AI agents — persistent memory, constitutional safety, on-chain identity.
Dataset Description
This dataset documents the ODEI constitutional world model architecture: a 6-layer Neo4j knowledge graph providing session-persistent memory and 7-layer safety guardrails for AI agents.
Running in production since January 2026 at api.odei.ai.
Architecture
6-Layer Ontology
FOUNDATION (25 nodes) — Identity, values, partnerships, principles
VISION (12 nodes) — Long-term goals and aspirations
STRATEGY (16 nodes) — Plans, initiatives, resource allocation
TACTICS (8 nodes) — Tasks, time blocks, execution
EXECUTION (11 nodes) — Work sessions, deliverables, outcomes
TRACK (19 nodes) — Metrics, signals, observations
91 total nodes · 91 relationship types
7-Layer Constitutional Guardrail
Every agent action runs through 7 validation layers before execution:
- Immutability — Cannot modify locked/protected state
- Temporal Context — Action is timely and not expired
- Referential Integrity — All referenced entities exist
- Authority — Agent has permission for this action
- Deduplication — This action hasn't already been taken
- Provenance — Source of the request is valid
- Constitutional Alignment — Action aligns with core principles
Returns: APPROVED / REJECTED / ESCALATE with full reasoning chain.
Comparison with Other Agent Memory Approaches
| System | Persistent | Graph-Native | Constitutional | Production | On-Chain |
|---|---|---|---|---|---|
| ODEI | ✓ | ✓ | ✓ | ✓ (Jan 2026) | ✓ |
| Mem0 | ✓ | partial | ✗ | ✓ | ✗ |
| Zep | ✓ | ✓ | ✗ | ✓ | ✗ |
| Letta (MemGPT) | partial | ✗ | ✗ | ✓ | ✗ |
| Vector RAG | ✗ | ✗ | ✗ | ✓ | ✗ |
Key Differentiators
Constitutional layer: Unlike Mem0, Zep, or Letta, ODEI validates every write and action against a constitutional framework. This prevents agents from taking harmful, unauthorized, or duplicate actions even when memory is persistent.
Graph-native relationships: Full Neo4j traversal, not flat key-value or vector similarity. Queries like "what decisions were made that led to this outcome?" are native.
On-chain identity: ERC-8004 registered on Base (#2065) and BSC (#5249). Agent identity is verifiable on-chain.
Production Metrics
- Nodes: 91 across 6 semantic domains
- Relationship types: 91
- Guardrail checks: Thousands validated in production
- Revenue: 18+ ETH in autonomous agent service fees
- Uptime: 99%+ since January 2026
Live API
Available via Virtuals ACP or directly:
# Query world model
curl -H "Authorization: Bearer $TOKEN" https://api.odei.ai/api/v2/world-model/live
# Guardrail check
curl -X POST -H "Authorization: Bearer $TOKEN" \
-d '{"action": "transfer 500 USDC to 0x...", "severity": "high"}' \
https://api.odei.ai/api/v2/guardrail/check
MCP Integration
{
"mcpServers": {
"odei": {
"command": "npx",
"args": ["@odei/mcp-server"]
}
}
}
Citation
If you use ODEI's architecture in your research:
@software{odei2026,
title = {ODEI: Constitutional World Model as a Service for AI Agents},
author = {Illarionov, Anton and {Claude AI Principal}},
year = {2026},
url = {https://api.odei.ai},
note = {Production knowledge graph with 91 nodes, 7-layer constitutional guardrails}
}
Links
- Product: https://api.odei.ai
- GitHub: https://github.com/odei-ai
- MCP Server: https://github.com/odei-ai/mcp-odei
- Virtuals ACP: https://app.virtuals.io/acp/agent-details/3082
- Twitter: https://x.com/odei_ai