Recce · Architecture

Technical
architecture
reference.

Component layers, integrations, and deployment topology for engineering and security teams evaluating Recce.

Request Demo View Security Fabric →
System Diagram

How Recce is structured.

Four control bands sit above the platform layer. A shared policy core connects them. Band D collects telemetry from all three and pushes to external SIEM.

AI Intake Recce Policy Core AI Workloads BAND A Model Supply Chain ModelScan · AIBOM · Cosign · OPA Trusted registry BAND B Data Ingestion ClamAV · PII scan · Signing Milvus vector gate BAND C AI Guardrails NeMo · OPA · Garak Prompt · API · Runtime BAND D Security Telemetry Audit log · Evidently AI SIEM / SOC export Recce POLICY CORE Policy-as-code (OPA) Identity & access control Cryptographic signing plane Audit event router → Band D Model serving Applications APIs & services Agentic systems Edge / on-prem telemetry → Band D → SIEM / SOC
Component Layers

Every component, where it fits.

Recce is composed of purpose-built components that replace or wrap existing open-source security tooling.

Band A — Model Supply Chain
ModelScan
CVE & vulnerability detection in model files
AIBOM Generator
AI Bill of Materials generation & attestation
Cosign / Sigstore
Cryptographic signing & verification
OPA Gate
Policy enforcement at promotion time
Model Registry
Trusted artifact store with provenance
Band B — Data Ingestion Security
ClamAV
Malware scanning on all ingested documents
PII Scanner
Sensitive data detection & field-level redaction
Data Signing
Per-asset signing before vector DB ingestion
Milvus Gate
Vector DB admission control for clean assets only
Band C — AI Guardrails
NeMo Guardrails
Prompt and output safety rail enforcement
OPA Runtime
Per-call policy engine for inference requests
Garak
Automated adversarial red-team probes
Tool Allow-List
Per-session agent tool scope enforcement
Output Classifier
Post-generation policy compliance filter
Band D — Security Telemetry
Audit Log
Immutable, tamper-evident event store
Evidently AI
Model drift & data quality monitoring
SIEM Exporter
CEF/LEEF formatted telemetry push
Compliance Reporter
Scorecard generation against mapped frameworks
Integrations

Works with your existing stack.

Recce integrates with the tools your security, platform, and AI teams already use — without requiring you to swap infrastructure.

Splunk
SIEM telemetry export in CEF format via HTTP Event Collector
Band D
Microsoft Sentinel
Log Analytics workspace ingestion via CEF/LEEF connector
Band D
IBM QRadar
Event log forwarding with LEEF format for AI security events
Band D
Elastic / ELK
Audit log shipping to Elasticsearch for custom SIEM builds
Band D
Milvus
Vector database integration for Band B security gate enforcement
Band B
Cosign / Sigstore
Cryptographic signing and verification for model artifacts
Band A
OPA (Open Policy Agent)
Policy-as-code enforcement across Bands A, B, and C
All Bands
NeMo Guardrails
Prompt and output safety policy enforcement (Band C)
Band C
NVIDIA NIM / Triton
Model serving integration for runtime guardrail attachment
Band C
Evidently AI
Model drift and data quality monitoring for Band D telemetry
Band D
ClamAV
Malware scanning engine for Band B document ingestion pipeline
Band B
Rancher / SUSE
Kubernetes platform integration — cluster-level deployment support
Platform
Get Started

Ready to evaluate the architecture?

Request a technical session with the Recce team. We'll walk through the full component diagram against your existing infrastructure.

Request Demo View Lifecycle →