RAG-Powered Security Intelligence

AI Security Analyst
that understands YOUR infrastructure

Ask security questions in natural language. Get answers grounded in your real infrastructure data — not generic advice. Powered by Retrieval-Augmented Generation (RAG) over your Neo4j attack graph, CVE database, and compliance scores.

Available on Pro and Enterprise plans · Powered by Gemini 1.5 Pro

What is RAG in Cybersecurity?

Retrieval-Augmented Generation (RAG) is an AI architecture that combines the reasoning power of Large Language Models (LLMs) with real-time data retrieval from your actual systems. In cybersecurity, this means the AI doesn't just rely on its training data — it actively queries your infrastructure graph, vulnerability database, and compliance scores before generating a response.

Traditional AI security tools either give generic advice based on training data, or require manual context — you have to copy-paste CVE lists, compliance reports, and asset inventories into a chat. RAG eliminates this: the AI automatically retrieves the relevant context from your environment in real-time.

EchelonGraph's RAG implementation is unique because it uses direct graph database queries (Neo4j Cypher) and structured SQL queries (PostgreSQL) instead of vector embeddings. This means zero hallucination risk on factual data — every CVE ID, asset name, and CVSS score in the response comes directly from your database, not from a fuzzy similarity match.

The RAG Pipeline

4 data sources, one intelligent response

Every query retrieves context from your real infrastructure data — not pre-trained knowledge.

🔗

Neo4j Attack Graph

Real-time infrastructure topology — assets, VPCs, subnets, security groups, internet-facing nodes, and blast radius attack paths.

47 nodes, 3 attack paths
🛡️

CVE Database

Thousands of vulnerabilities with CVSS scores, severity ratings, exploit availability, and asset-to-CVE mapping.

Real-time NVD feed
📋

Compliance Scores

17 frameworks — SOC 2, NIST 800-53, CIS AWS/GCP/Azure/K8s, Pod Security Standards, PCI-DSS, HIPAA, ISO 27001, GDPR, plus 5 AI-specific (NIST AI-RMF, EU AI Act, ISO 42001, MITRE ATLAS, OWASP LLM Top 10) — with per-control pass/fail status.

100+ controls scored

Risk Factors

Weighted risk scoring across 5 dimensions: CVE exposure, misconfigurations, compliance gaps, blast radius, and SLA breaches.

Letter grade A+ to F
Real Examples

Ask anything about your security

What are my most critical vulnerabilities?
Your top risk is CVE-2026-34160 on prod-db-01 (CVSS 9.8). It's internet-facing via web-proxy-01 with blast radius of 12 downstream services. Patch immediately — public exploit exists.
🔗 Graph🛡️ CVEs⚡ Risk
Am I SOC 2 compliant?
Your SOC 2 coverage is 73% (58/80 controls passing). The 22 failing controls are concentrated in CC6.1 (Logical Access) and CC7.2 (Monitoring). Key gaps: unpatched RDS instances and missing MFA on 3 IAM roles.
📋 Compliance🔗 Graph
What would happen if web-proxy-01 was compromised?
Blast radius analysis shows 3 attack paths from web-proxy-01 → prod-db-01 (sensitive). Cumulative CVSS: 18.4. 12 downstream services affected including 2 data stores with PII. The lateral movement path goes through sg-internal (port 5432 open).
🔗 Graph🛡️ CVEs⚡ Risk
Give me a security briefing for my board meeting
Risk Score: 67/100 (Grade D — 'Poor'). 12 critical findings open, 28 high. SOC 2 at 73%, HIPAA at 81%. Biggest concern: prod-db-01 is internet-reachable with 2 unpatched CRITICAL CVEs. Recommend emergency patching within 24h.
⚡ Risk📋 Compliance🛡️ CVEs🔗 Graph
🔒
Your data stays yours

Each answer is locked to your account.

The AI Analyst only reads from your environment. We never mix your data with another customer's, never train on it, and never let the model wander outside what you've asked about. Audit-ready logs are scoped to your team alone.

Why EchelonGraph RAG

EchelonGraph RAG vs. Generic AI

FeatureEchelonGraph RAGGeneric AI Chatbots
Data SourceYour actual Neo4j graph + PostgreSQLGeneric training data
AccuracyReferences real CVE IDs, asset namesMay hallucinate
Retrieval MethodDirect graph + SQL queriesVector similarity search
FreshnessReal-time (queries live DB)Stale embeddings
ContextTenant-specific infrastructureShared/generic knowledge

How EchelonGraph's RAG Pipeline Works

01

Intent Classification

Your question is classified into intent categories — CVE/vulnerability, compliance, risk, or general. This determines which databases to query, avoiding unnecessary round-trips.

02

Context Retrieval (RAG)

Based on intent, the system queries your Neo4j attack graph (Cypher queries for blast radius and topology), PostgreSQL (findings, CVEs, compliance scores), and risk scoring engine — all scoped to YOUR tenant.

03

Context Assembly

Retrieved data is assembled into a structured context document: infrastructure summary, top riskiest assets, critical attack paths, open findings, compliance gaps, and risk factor breakdown.

04

LLM Generation

The context + your question + a security-expert system prompt are sent to Gemini 1.5 Pro (temperature 0.1 for deterministic output). The model is instructed to ONLY reference data from the context — never hallucinate.

05

Source Attribution

Every response includes source badges showing exactly which data sources were used — so you know whether the answer came from your Graph, Findings, Compliance scores, or Risk analysis.

Ready to try RAG-powered security?

Start asking your AI Security Analyst questions today. Get intelligence grounded in YOUR infrastructure data — not generic advice.

AI Analyst available on Pro & Enterprise plans · No credit card required