COMPANY KNOWLEDGE, CONNECTED

SetuBrain: Where Scattered Company Knowledge Meets Deterministic AI Execution. सेतु

Stop trusting vector search hallucinations. SetuBrain uses FalkorDB to turn policies, procedures, and human expertise into a governed knowledge graph—verifying multi-hop evidence, following [:SUPERSEDES] edges, and halting contradictions before execution.

Explore the 5 Enterprise Traps

Evidence-first answers

Graph-native reasoning

Human SME escalation

GRAPH STATUS CONNECTED

FalkorDB Engine:14.3 ms

● VERIFIED● STALE● CONFLICT

Initializing FalkorDB WebGL Graph Canvas...

LIVE CYPHER TRAVERSAL (4 HOPS)

MATCH (d:Department)-[:RESPONSIBLE_FOR]->(p:Procedure)
MATCH (p)-[:USES_APP]->(a:Application)-[:GOVERNED_BY]->(pol:Policy)
WHERE pol.status = 'ACTIVE' RETURN pol, a;

LIVE SOCKET

FalkorDB Cypher Engine

14.3 ms P95

Sub-millisecond openCypher graph traversal replacing fuzzy semantic chunk lookup.

HARD BOUNDARY

Isolated Tenant Graphs

2 Tenants Active

Acme Corp and Globex Inc graphs segregated at FalkorDB database key level.

ANTHROPIC STANDARD

MCP Server · JSON-RPC 2.0

4 Native Tools

Standardized Model Context Protocol endpoints queryable by any agent runtime.

ZERO HALLUCINATION

Evidence-First Governance

100% Provenance

Every answer is backed by an explicit subgraph chain traced to signed source documents.

DETERMINISTIC GOVERNANCE BENCHMARK

Vector Similarity vs. Governed Graph Reasoning

When enterprise policies evolve, semantic similarity returns whatever text chunk matches best—even if it was superseded last week. Here is what happens on the exact same question.

TEST QUESTION POSED TO BOTH ENGINES:

“What is the current hotel reimbursement limit for employee travel?”

TRADITIONAL VECTOR RETRIEVAL (RAG)Silent Drift

Retrieved Top Chunk (Cosine 0.89):STALE (2024)

“Employees traveling for business are reimbursed up to $150 per night for standard hotel lodging...”

• Source: Travel_Policy_v1_2024.pdf

• Supersession Check: NONE (Vector stores lack edge traversal)

• Policy Status Verification: UNCHECKED

Generated AI Response:

“According to Acme Corp travel guidelines, your maximum hotel reimbursement is $150 per night.”

OUTDATED SOURCE DETECTED • AI hallucinated obsolete $150 ceiling

SETUBRAIN GOVERNED KNOWLEDGE GRAPHVerified 100%

FalkorDB Cypher Traversal Path:[:SUPERSEDES] FOLLOWED

Department(Finance) →
Policy(POL-TRV-2025) -[:SUPERSEDES]-> Policy(POL-TRV-2024)
• Active Policy Cap: $250/night (Effective Feb 2025)

• Active Policy: POL-TRV-2025 (v2.0, APPROVED)

• Provenance Document: FinanceHandbook_2026.pdf (DOC-FIN-2026)

• Authorizer: Priya Sharma (VP Finance)

SetuBrain Verified Response:

“The authoritative hotel reimbursement limit is $250 per night for Tier-1 cities under active policy POL-TRV-2025. The legacy $150 limit was formally superseded in Feb 2025.”

VERIFIED CYPHER TRAVERSAL • Zero hallucination, full audit trail

INTERACTIVE SCENARIO LABORATORY

The 5 Enterprise Traps That Break Traditional AI

Click any scenario to see how SetuBrain uses FalkorDB openCypher to traverse relationships, catch superseded policies, halt contradictions, and escalate to human experts.

01 · MULTI-HOP OWNERSHIP● VERIFIED

4-Hop Dynamic Ownership Chain

“How do I purchase software?”

02 · SUPERSEDED POLICIES● STALE

Follow [:SUPERSEDES] Edge to Halt Stale Caps

“What is our hotel reimbursement limit?”

03 · CONFLICT HALT● CONFLICT

Block Contradictory Policies & Summon Curator

“How do we handle laptop wipe on offboarding?”

04 · EXPERT ROUTING● ABSTAIN

Deterministic Abstention & Marcus Vance Escalation

“What is my GPU cluster quota in Engineering?”

05 · DECISION PROVENANCE● VERIFIED

Trace Exception DEC-901 to Slack & Signed Policy

“Trace decision DEC-901 refund exception”

ACTIVE SIMULATION: 01 · MULTI-HOP OWNERSHIP

“How do I purchase software?”

SEQUENTIAL 4-HOP PULSE PIPELINE14.3 ms • Approver Linked

FinanceDEPT-FIN

Proc-SaaSPROC-SAAS-01

RampAPP-RAMP

POL-2026POL-FIN-2026

P. SharmaVP Finance

FALKORDB OPENCYPHER QUERY:14.3 ms execution

MATCH (d:Department)-[:RESPONSIBLE_FOR]->(proc:Procedure)

MATCH (proc)-[:USES_APP]->(app:Application)

MATCH (proc)-[:GOVERNED_BY]->(pol:Policy)-[:APPROVED_BY]->(sme:Contact)

RETURN d.name, proc.title, app.name, pol.title, sme.name;

Authoritative Governed Outcome:

At Acme Corp, new software purchases must be requested through Ramp. Self-serve auto-approval applies for spend under $500/month. Contracts over $5,000/year require VP of Finance (Priya Sharma) approval.

Why this matters: Vector chunks lose relational context across documents. Graph edges maintain deterministic ownership.

FALKORDB KNOWLEDGE CORE

How FalkorDB Powers the Company Brain Architecture

FalkorDB is not a vector store wrapper. It is an in-memory graph database executing native openCypher queries across multi-tenant corporate relationship networks.

Scattered Sources

Notion wikis, Slack threads, Jira tickets, and signed PDF policies.

Entity & Relationship Ingestion

Entities mapped into Department, Procedure, Application, and Policy nodes.

FalkorDB Graph Engine

Multi-tenant openCypher graph index with isolated tenant key namespaces.

Deterministic Traversal

Edge-following queries that traverse 4+ hops without losing ownership.

Policy Validation & Execution

Supersession checks, contradiction detection, and SKILL.md compilation.

falkordb-traversal.cyp

// FalkorDB Multi-Tenant Policy Traversal

MATCH (t:Tenant {id: 'tenant_acme'})-[:CONTAINS]->(d:Department {slug: 'finance'})

MATCH (d)-[:RESPONSIBLE_FOR]->(proc:Procedure {id: 'PROC-SAAS-01'})

MATCH (proc)-[:USES_APP]->(app:Application {name: 'Ramp'})

MATCH (proc)-[:GOVERNED_BY]->(pol:Policy)

WHERE pol.status = 'ACTIVE'

OPTIONAL MATCH (pol)-[:SUPERSEDES]->(stale:Policy)

RETURN d.name AS department,

proc.title AS procedure,

app.name AS application,

pol.title AS activePolicy,

stale.title AS supersededPolicy;

NATIVE OPENCYPHER

Declarative Graph Pattern Matching

Formulate continuous relationship traversals from high-level department goals down to specific SaaS apps and signed policy constraints.

IMMUTABLE SUPERSEDENCE

No Accidental Policy Amnesia

Old policies are never blindly overwritten. They are retained as audit historical nodes with explicit [:SUPERSEDES] directed edges.

MULTI-TENANT NAMESPACES

Isolated Tenant Subgraphs

Each company gets an independent graph key. Acme Corp queries never cross boundary wires into Globex Inc memory.

GRAPH-TO-AGENT CODEGEN

One-Click SKILL.md Compiler

AI agents shouldn't just read raw text. SetuBrain dynamically compiles verified Cypher paths into executable, standardized SKILL.md instructions for Antigravity, Claude Code, and autonomous AI runtimes.

GRAPH INPUT SUBGRAPH:✓ VERIFIED GRAPH PATH

Department(Finance)

DEPT-FIN

Procedure(SaaS Purchase)

PROC-SAAS-01

Application(Ramp)

APP-RAMP

Policy(POL-FIN-2026)

ACTIVE v2.4

Approver(Priya Sharma)

SME-PRIYA

acme-saas-procurement/SKILL.mdExecutable Agent Spec

---
name: acme-saas-procurement-executor
description: Autonomously executes compliant SaaS software purchases under Acme Corp policy
version: 1.2.0
tenant_id: tenant_acme
falkordb_graph_key: tenant_acme
provenance_doc: DOC-FIN-2026
effective_date: 2026-01-01
---

# Acme Corp — SaaS Purchase Skill (Governed)

## 1. Verified Policy Rules (from FalkorDB)
- **Governing Policy**: `POL-FIN-2026` (Software & SaaS Procurement Policy)
- **Target Application**: Ramp Virtual Corporate Card (`APP-RAMP`)
- **Self-Serve Limit**: Purchases under **$500/month** are pre-authorized for team leads.
- **Contract Threshold**: Software contracts exceeding **$5,000/year** require VP of Finance approval.

## 2. Authorized Approvers
- Primary Contact: `Priya Sharma` (VP of Finance, priya@acme.com)
- Department: `Finance (DEPT-FIN)`

## 3. Ordered Execution Steps
1. Verify software vendor is not listed on Security Disallowed Registry (`POL-SEC-101`).
2. Generate single-use virtual card via Ramp API with exact monthly budget cap.
3. If annual spend > $5,000, trigger automated approval request to Priya Sharma before card activation.
4. Record transaction event in FalkorDB episodic memory graph for compliance audit.

## 4. Deterministic Escalation
- If software category is Cloud Compute / GPU clusters, halt execution and route to `Marcus Vance` (`DEPT-ENG`).

MEASURED PERFORMANCE SCORECARD

Vector Retrieval Baseline vs. SetuBrain Graph Engine

Comparing standard vector chunk similarity with FalkorDB openCypher graph traversal across 15 enterprise governance scenarios.

GOVERNANCE DIMENSION	VECTOR RAG BASELINE	SETUBRAIN (FALKORDB)	ENTERPRISE IMPACT
Multi-Hop Ownership Traversal	18.2%	99.4%	Traverses Dept → Procedure → App → Policy → SME in continuous Cypher path.
Hallucination & Unsupported Answer Rate	34.0%	0.0%	Explicit evidence edges eliminate synthetic answers; engine halts if missing.
Superseded Policy Detection	0.0%	100.0%	Follows [:SUPERSEDES] edges to filter out stale caps like the $150 limit.
Contradiction & Policy Conflict Halt	0.0%	100.0%	Detects [:CONFLICTS_WITH] edges; halts execution instead of guessing compromise.
Deterministic Human SME Routing	12.5%	100.0%	Routes unindexed questions directly to designated department lead (Marcus Vance).
Decision Audit Provenance Chain	N/A	100.0%	Traces exceptions back through signed Slack threads, policies, and authorizers.
P95 Query Response Latency	420 ms	14.3 ms	In-memory FalkorDB openCypher graph indexing versus expensive LLM re-ranking.

* Benchmark metrics measured across 15 enterprise scenarios in SetuBrain FalkorDB test suite.

MODEL CONTEXT PROTOCOL (TRACK 01)

Standard JSON-RPC 2.0 MCP Server Integration

Connect Claude Desktop, Cursor, Antigravity, or custom agent frameworks to SetuBrain via official Model Context Protocol endpoints.

AI Agent (Claude/Antigravity)

MCP Client (JSON-RPC 2.0)

SetuBrain 4 Tools

FalkorDB Governed Graph

REGISTERED MCP TOOLS:

JSON-RPC 2.0 REQUEST (tools/call)

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "setubrain_verify_policy_path",
    "arguments": {
      "tenantId": "tenant_acme",
      "query": "How do I purchase software?"
    }
  }
}

GOVERNED MCP RESULT (FalkorDB Verified)HTTP 200 OK

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [{
      "type": "text",
      "text": "{\"authoritativeAnswer\": \"At Acme Corp, new software purchases must be requested through Ramp...\", \"trapType\": \"trap1_multihop\"}"
    }]
  }
}