SEOCHO | The OS for Agentic Knowledge Graphs
The Operating System for
Agentic Knowledge Graphs
1. Extract real-world data into a semantic graph governed by explicit SHACL rules.
2. Orchestrate dedicated agents to debate precisely over isolated data subgraphs.
3. Guarantee 100% routing transparency and execution provenance via Opik.
from seocho import CoreEngine, GraphNode
# 1. Initialize the engine with graph backend
engine = CoreEngine(
graph_uri="bolt://neo4j:7687",
ontology_path="./ontology.ttl"
)
# 2. Define semantic routing rule
engine.add_route(
name="compliance_check",
condition=lambda q: "regulation" in q.intent,
target_graph="fss_regulatory_graph"
)
# 3. Create persistent agent over the graph
agent = engine.create_agent(
name="ComplianceOfficer",
system_prompt="You enforce internal philosophy using semantic trace DAGs.",
memory_type="infinite_graph"
)
# Execute
response = agent.run("Does the new onboarding flow violate FSS rule 42?");
print(response.decision) # -> CONDITIONAL_GO
Everything you need to build
agentic knowledge graphs.
Not just RAG. A complete framework for semantic orchestration, continuous evaluation, and deterministic routing.
Infinite Memory
Agents store episodic and semantic memory directly in Neo4j. Context window limits are obsolete.
Semantic Debate Engine
Multiple LLM agents (Generator, Verifier) debate complex queries against the topology contract before returning answers. Guaranteed hallucination reduction.
Deterministic Routing
Query surfaces are policy-gated. A semantic router maps queries to the exact graph workspace instantly. No heuristic lineage reconstruction.
Enterprise Governance
Ontology deployment paths are offline-heavy and rollback-safe via SHACL rules.
Quickstart
Get up and running locally in under 5 minutes.
1 Clone & Configure
cd seocho && cp .env.example .env
2 Spin Up Services
3 Execute Debate Query
System Architecture
From document ingestion to asynchronous agent debate routing.
graph TD
classDef external fill:transparent,stroke:#52525B,stroke-width:1px,stroke-dasharray:5 5,color:#A1A1AA
classDef pipeline fill:#ffffff10,stroke:#d4d4d8,stroke-width:1px,color:#e4e4e7
classDef db fill:#ffffff05,stroke:#a1a1aa,stroke-width:1px,color:#d4d4d8
classDef agent fill:#ffffff15,stroke:#f4f4f5,stroke-width:1px,color:#ffffff
classDef user fill:transparent,stroke:none,color:#E4E4E7,font-weight:bold
User(("User Query")):::user
subgraph DataPlane["Data Extraction Pipeline"]
DS["Raw Sources"]:::external-->Extractor["LLM Extractor"]:::pipeline
Extractor-->Linker["Entity Linker & Semantic Deduplication"]:::pipeline
Linker-->DBManager["Dynamic DB Provisioner"]:::pipeline
end
subgraph ControlPlane["Parallel Debate Orchestrator"]
Router["Debate Router"]:::agent
Agent1["Agent (Domain A)"]:::agent
Agent2["Agent (Domain B)"]:::agent
Supervisor["Synthesis Supervisor"]:::agent
end
Neo4j[("Neo4j / DozerDB Partitions")]:::db
DBManager --> Neo4j
User --> Router
Router -->|Fan-Out Async| Agent1
Router -->|Fan-Out Async| Agent2
Agent1 -.-> Neo4j
Agent2 -.-> Neo4j
Agent1 -->|Collect State| Supervisor
Agent2 -->|Collect State| Supervisor
Supervisor -->|Resolved Answer| User
The SEOCHO Blog
Concepts, reasoning, and major release announcements.
Image Prompts for Landing Page
Discover the latest releases and updates from the SEOCHO engineering team.
SEOCHO Runtime Update: Strict E2E Smoke Tests and the Open Source Playbook
Discover the latest releases and updates from the SEOCHO engineering team.
Feasibility Review Framework & Rubrics
Multi-role feasibility review framework and Go/No-Go rubric for graph data implementations.
Latest Updates
Continuously integrated signals from GitHub Releases.
- 2026-02-21 feat: implement P1 readiness routing, health split, and elementId migration #033c6c5
- 2026-02-21 fix: add agent sdk adapter and debate readiness contract #44fb77a
- 2026-02-21 docs: clarify quickstart for custom-port environments #cbf3e8f
- 2026-02-21 docs: add user-first quickstart and architecture priority docs #7f968fd
- 2026-02-21 Fix debate runtime compatibility and skip missing databases #a3b7b7f
FAQ
What exactly is Parallel Debate? ↓
It's an orchestration framework where multiple isolated LLM agents execute reasoning paths concurrently against the graph database. A Supervisor agent then evaluates conflicting conclusions from these agents to provide the single most grounded answer.
Why do I need dynamic databases? ↓
Unlike monolithic deployment strategies, SEOCHO provisions entirely isolated Neo4j/DozerDB instances for distinctly separate datasets on the fly. This enforces hardened physical boundaries and protects schema integrity per ingestion cluster.
Does it integrate with my existing Neo4j? ↓
Yes. While we default to containerized DozerDB endpoints, SEOCHO's Data Pipeline can map semantic entities and perform insertions into any standard Neo4j or Aura graph engine via the Bolt protocol specifications.
What observability tools are used? ↓
We implement native integration hooks with Opik. This provides millisecond tracing of prompt payloads, LLM token expenditures, latency waterfalls, and a full structural view into the reasoning tree of the parallel agents.