Keep your rule engine.
Cover the blind spots with AI
An AI verdict layer that attaches to the else branch of the rule-based inspection systems you already operate — one API call, zero rebuild. Only the edge cases your rules can't express get delegated to AI.
Architecture
A non-invasive verdict layer
that augments your legacy engine
Proven rule engines — ERP SoD, network IDS, asset patch checks — stay exactly as they are. Only undefined edge cases are delegated to an external verdict service. No rule-catalog migration, no data migration.
Rip-and-replace vs Crux
Verdict Pipeline
How a verdict flows
When an edge case hits your else branch, one API call requests a verdict — returned instantly at high confidence, or escalated to the human review queue when confidence is low.
Key Features
Key Features
Non-Invasive Integration
Achieve next-generation AI capabilities with zero structural disruption and zero data migration — just one API call added to your else branch.
No rule-catalog migration, no data migration
Intelligent Verdict Routing
Deterministic conditions run as RULE, context-dependent judgments as AGENT, and borderline cases as HYBRID — the right method per case.
Deterministic and probabilistic verdicts, combined
Confidence-Based Escalation
High-confidence verdicts return immediately. Low-confidence verdicts trigger an automated escalation to the human-in-the-loop governance queue.
Returns verdict + confidence + rationale
Approval-Based Governance
As verdict history accumulates, AI recommends reclassifications and humans approve them. Control always stays with your team.
Built for EU AI Act human-oversight requirements
BYOM Architecture
Connect endpoints you already trust — Azure OpenAI, AWS Bedrock, or on-premises models — to ensure strict data sovereignty.
Managed LLM option available
Audit Trail
Every verdict execution and approval decision is tracked per tenant, with full rationale recorded for regulatory audits.
Fully tenant-isolated audit logs
Use Cases
Turn the risks hiding between rules
into verdict targets
Cover the cases that fall outside judgment entirely because if-else can't express them.
Coverage Gaps
Cases your rule matrix has no concept for — like delegated ERP approval authority missing from the SoD matrix — judged in context.
Composite Risks
Patched, authorized, untouched for 90 days — every rule passes green individually, but the combination signals a neglected-asset risk.
Context-Dependent Cases
Macro execution intent, anomalous movement across sessions — conditions no numeric threshold can express.
Regulatory Interpretation
Checklist clauses run as RULE; clauses needing interpretation are judged with RAG retrieval, citing regulation text verbatim.
Integration
Plugs into how you already operate
From the verdict API to approval workflows, Crux connects to the tools your teams already use.
REST API
A single verdict endpoint called from your else branch
BYOM Endpoints
Azure OpenAI · AWS Bedrock · on-premises models
Approval Workflow
AI recommendations surfaced in your existing approval system
Audit Log Export
Verdict and approval history delivered to SIEM and audit systems
Don't replace your rule engine.
Augment it
Start covering your blind spots with the inspection systems you run today.