Legal & compliance

Our security program operates under AICPA trust services criteria across Security, Availability, and Confidentiality. A formal nRouter SOC 2 Type II observation window is currently in progress with independent auditors, while underlying cloud hosting infrastructure (Azure, Supabase) is fully SOC 2 Type II certified.

• Automated continuous monitoring tracking cloud configurations, IAM policies, and cryptographic standards in real time.
• Role-based access control (RBAC), multi-factor authentication (MFA) enforcement on all systems, and automated quarterly access certification.
• Reports are available to enterprise prospects and procurement teams under mutual non-disclosure agreement (NDA) via the Trust Center.

Full compliance with the European General Data Protection Regulation (Regulation (EU) 2016/679). nRouter serves as a data processor for inference telemetry and customer payloads, providing a formal Data Processing Agreement incorporating European Commission Standard Contractual Clauses (SCCs).

• Incorporates Module 2 (Controller-to-Processor) and Module 3 (Processor-to-Processor) Standard Contractual Clauses for legal international transfers.
• Automated workflows for Data Subject Requests (DSR) honoring access, rectification, erasure (Right to be Forgotten), and data portability.
• Designated Data Protection Officer (DPO) and formal 72-hour regulatory breach notification procedures under GDPR Article 33.

Enterprise healthcare and life sciences organizations can safely process Protected Health Information (PHI) through nRouter. We execute formal Business Associate Agreements (BAAs) defining mutual security obligations under 45 CFR Part 160 and Part 164 Subparts A and C.

• Zero Data Retention (ZDR) configuration permanently disables payload logging to ensure no PHI is stored on persistent storage disks.
• Strict technical safeguards: AES-256-GCM encryption at rest, TLS 1.3 in transit, and isolated tenant cryptographic contexts.
• Comprehensive workforce compliance training, security background checks, and documented incident response runbooks.

Our Information Security Management System (ISMS) is modeled in strict alignment with ISO/IEC 27001:2022 specifications, providing a systematic approach to risk management, threat mitigation, vulnerability management, and business continuity.

• Formal risk assessment methodology updated quarterly to evaluate newly emerging generative AI threat vectors and LLM vulnerabilities.
• Automated dependency scanning, container vulnerability scans, static code analysis (SAST), and annual third-party penetration tests.
• Comprehensive business continuity management with recovery point objectives (RPO < 1 min) and recovery time objectives (RTO < 15 min).

We provide an absolute contractual guarantee that neither customer prompts, model outputs, embeddings, nor fine-tuning datasets are ever used to train, retrain, or improve foundational AI models. This guarantee binds nRouter and our upstream model providers (OpenAI, Anthropic, Google Cloud, AWS Bedrock).

Our enterprise commercial agreements legally enforce zero-training commitments across every commercial API tier and endpoint. No customer conversation history is ever donated to public or private model weights.

All inference payload streams are proxied entirely in volatile RAM buffers. As soon as stream chunks are delivered to the client socket, buffer memory is immediately zeroed and reclaimed. No prompts, completions, or image payloads are ever written to disk or persistent storage databases.

Gateway spend logs retain strictly 22 non-content operational dimensions (request ID, tenant ID, virtual key ID, model identifier, provider, token counts, cost breakdown, latency, HTTP status code) for financial reconciliation and audit compliance.

Tenant isolation is enforced across both data storage and runtime layers. Our PostgreSQL control plane enforces Row-Level Security (RLS) with mandatory tenant pinning on every query, connecting via transactional poolers configured with strict NOBYPASSRLS privileges to prevent cross-tenant data leakage.

Each organization's virtual API keys, routing configurations, spending budgets, and rate-limit buckets operate in isolated cryptographic namespaces. Gateway sidecar compute services operate under mutual TLS 1.3 with air-gapped network boundaries and zero direct database connectivity.

To satisfy regional sovereignty and regulatory requirements (such as EU Data Boundary and US federal mandates), nRouter supports geofenced inference routing policies. Customers can configure strict US-only, EU-only, or APAC-only egress rules.

When regional geofencing is activated, our routing engine restricts provider endpoint selection to deployments physically located within the designated ISO 3166-1 territory, preventing cross-border hops and ensuring compliance with local data sovereignty laws.

Organizations maintain complete ownership and responsibility for securing their virtual API keys and administrative credentials. We provide granular tools to enforce least-privilege security, including per-key spending ceilings, model access whitelists, team-scoped access controls, and automated credential rotation.

If an organization suspects that an API key has been exposed or compromised, it must immediately revoke the key through the dashboard or API to halt egress.

Our high-availability gateway infrastructure is engineered for resilience, running across multiple geographic regions with automated health checks, provider circuit breakers, and sub-millisecond route failover.

Eligible Enterprise plans include a 99.99% monthly availability Service Level Agreement (SLA). In the event of unplanned downtime exceeding agreed error budgets, customers receive transparent financial service credits applied to their monthly billing invoice.

nRouter enforces fair resource allocation and strictly prohibits harmful activities, including the generation of unlawful content, malicious prompt injection attacks against third parties, denial-of-service attempts, unauthorized penetration testing, or attempts to circumvent gateway metering.

Our edge WAF and preflight inspection layers enforce rate limiting (RPM/TPM) and guardrail scanning to protect overall system health and guarantee fair access for all tenants.

Models are offered at exact provider list prices with zero per-token markup. All billing follows a transparent two-tier funding structure: inference requests draw against your monthly subscription plan allowance first, overflowing to prepaid top-up credits only when extra usage is explicitly enabled.

Before upstream egress, a downward-adjusting credit reservation hold is placed based on estimated tokens. Upon stream completion, spend is settled to the exact millicent of list price, and any unused reservation hold is immediately released.