ZERO-EGRESS GATEWAY
Metriota Trust & Security Center

Zero-Trust Governance for
Regulated Enterprise AI.

Security, privacy, and compliance are built directly into the Metriota Gateway architecture[cite: 1]—deployable entirely inside your own VPC or air-gapped perimeter so sensitive prompts, vault keys, and audit logs never leave your control[cite: 3].

Four Pillars of AI Perimeter Defense

Engineered to protect proprietary data, patient records, financial intelligence, and classified workloads without slowing down engineering velocity.

Reversible Zero-Egress Data Vault

Sensitive entities—including PII, PHI, MNPI, API credentials, and custom dictionary terms—are intercepted in-memory and replaced with synthetic tokens before leaving your network[cite: 3]. External model providers never see raw sensitive data, while authorized internal apps receive re-hydrated responses[cite: 3].

Inline Prompt Injection & RAG Firewall

Every user prompt and external RAG document is scanned in real time for adversarial jailbreaks, indirect prompt injections, and system-prompt extraction attempts before reaching your inference layer or downstream tool-calling agents[cite: 3].

Cryptographic Audit & WORM Lineage

Metriota emits structured, tamper-evident OpenTelemetry (OTLP) traces capturing policy evaluations, redaction events, model versions, and routing decisions—exportable as SHA-256 verified bundles for immutable Write-Once-Read-Many (WORM) compliance[cite: 3].

Hardware-Aware Sovereign Routing

Enforce strict data residency boundaries by classification level. Route general workloads to commercial cloud APIs while automatically localizing European citizen data[cite: 1] or locking restricted queries onto self-hosted vLLM or Ollama GPU nodes inside your VPC.

Data Flow & Threat Model

How Data Moves Through the Gateway

Every inference request passes through four deterministic isolation stages inside your perimeter before any external network call is permitted[cite: 3].

STAGE 01

Ingress Sanitization

Payloads and retrieved RAG chunks are inspected inline for jailbreaks, system-prompt overrides, and exfiltration payloads[cite: 3]. Malicious inputs fail closed (HTTP 403) before touching model compute[cite: 3].

MITIGATES: Prompt Injection & RAG Poisoning
STAGE 02

In-Memory Tokenization

Presidio NER and deterministic regex rules extract PII, PHI, MNPI, and custom secrets[cite: 3]. Sensitive spans are replaced with synthetic placeholders while encrypted mappings stay in ephemeral Redis with strict TTLs[cite: 1, 3].

MITIGATES: Third-Party API Data Egress
STAGE 03

Policy-Bound Routing

Classification rules enforce destination boundaries: restricted workloads stay pinned to local vLLM/Ollama nodes, while sanitized prompts can safely burst to approved cloud endpoints[cite: 3].

MITIGATES: Shadow AI & Residency Violations
STAGE 04

Re-Hydration & OTLP Trace

Synthetic tokens in the model response are re-hydrated inside your VPC for the authorized caller[cite: 3]. Only metadata spans (or SHA-256 hashed audit bundles when enabled) are emitted[cite: 3].

MITIGATES: Unaudited Black-Box Decisions
No Persistent Logs by Default: Unless explicitly configured for compliance audit trails, raw prompt text is never written to disk[cite: 1].
Daily CVE Container Scanning: Production images are scanned daily against CVE databases prior to release[cite: 1].
Responsible Disclosure: Report security findings directly to security@metriota.com under our vulnerability disclosure program[cite: 1].
Deployment Governance

Self-Hosted Shared-Responsibility Matrix

Because Metriota runs inside your private cloud or on-premise infrastructure, your organization retains full custody of data, keys, and network boundaries—meaning customer data never enters a third-party SaaS processor scope.

security@metriota.com
Control DomainWhat Metriota ProvidesWhat Your Team Controls
Gateway Software & Detection EngineHardened Docker images, Presidio NER/regex pipelines, prompt injection scanner, signed releases, and daily CVE vulnerability scanning[cite: 1, 3].Applying container version updates and pinning approved image digests in internal registries.
Compute, Network & PerimeterReference Docker Compose / Kubernetes manifests, localhost port-binding defaults (127.0.0.1), and zero-egress architecture guides.Cloud VPC / on-premise server hardening, TLS termination, firewall rules, and private network routing.
Vault Encryption & Key CustodyIn-memory tokenization, AES-256 vault payload encryption, and strict Redis Time-To-Live (TTL) auto-purge policies[cite: 1].Managing KMS/master encryption keys, Redis persistence policies, and volume-level disk encryption.
Identity & Access GovernanceRole-Based Access Control (RBAC), scoped gateway API keys, and OIDC/SSO integration hooks.Identity Provider (IdP) user provisioning, MFA enforcement, and periodic access reviews.
Audit Evidence & RetentionStructured OpenTelemetry (OTLP) spans, SHA-256 verified WORM evidence pack exports, and no-disk-logging defaults[cite: 1, 3].Configuring immutable WORM storage buckets (e.g., S3 Object Lock), SIEM ingestion, and regulatory retention schedules.
Formal Audit & Regulatory ScopeTechnical control mapping, architecture documentation, threat model, and enterprise BAA/DPA templates[cite: 1].Incorporating the self-hosted gateway into the organization's own SOC 2, ISO 27001, HIPAA, or SEC 17a-4 audit boundary.

Regulatory & Compliance Alignment

Self-hosted deployments include technical controls, audit logging, and configuration guidance designed to support your organization's compliance programs across regulated sectors.

Enterprise & GDPR

  • SOC 2 & ISO 27001 Aligned Controls
  • EU Data Localization (GDPR)[cite: 1]
  • Zero-Egress VPC Deployment
  • SAML 2.0 SSO & RBAC

Financial Services

  • SEC Rule 17a-4 (WORM Export)[cite: 3]
  • OCC SR 11-7 MRM Lineage[cite: 3]
  • MNPI & Deal-Code Vaulting[cite: 3]
  • FINRA Audit Evidence Packs[cite: 3]

Healthcare & Life Sci

  • HIPAA / HITECH Safeguards
  • 18-Identifier Safe Harbor PHI
  • Enterprise BAA Templates[cite: 1]
  • Clinical IP & Trial Isolation

Manufacturing & Defense

  • ITAR Egress Prevention
  • CMMC 2.0 Audit Evidence
  • CAD & Formula Tokenization
  • Air-Gapped Local Routing

Security & Procurement FAQ

Where are reversible vault tokens and encryption keys stored?
In self-hosted VPC deployments, all token mappings reside exclusively in your own Redis cluster with strict TTLs and AES-256 encryption[cite: 1]. Neither Metriota nor upstream LLM providers ever have access to the re-hydration lookup table[cite: 3].
Can Metriota run in a completely air-gapped environment?
Yes. All NLP/NER detection models, policy engines, databases, and observability collectors are packaged inside self-contained Docker images requiring zero outbound internet access when paired with internal inference servers (such as vLLM or Ollama)[cite: 3].
How do we review your security architecture, data handling, or BAA/DPA terms?
Because Metriota deploys directly inside your own VPC or air-gapped infrastructure, sensitive prompts and vault keys never leave your controlled environment[cite: 3]. Contact our team via the Trust Center request button or email security@metriota.com to request architecture diagrams, deployment hardening guides, and enterprise BAA/DPA templates[cite: 1].

Ready for a Security & Architecture Walkthrough?

Review our Docker Compose and Kubernetes deployment manifests, custom vault dictionaries, and WORM audit exports with our engineering team.