Scrutari
NIST FIPS 203 + 204 ready · ML-KEM-768 hybrid

Hybrid post-quantum TLS termination, in front of the backend you already run.

A hermetic Rust gateway terminating ML-KEM-768 + X25519 hybrid handshakes on your hostname. TLS 1.3 only, no classical-only fallback, AEAD ciphers only. Onboarding is two DNS records; certificate issuance and rotation run inside the gateway. Backend code unchanged.

Hermetic Rust runtime · TLS 1.3 only · No classical-only fallback · Single-process attack surface

Built on standards that hold up under audit.

  • NIST FIPS 203ML-KEM
  • NIST FIPS 204ML-DSA
  • NIST FIPS 205SLH-DSA
  • SOC 2 / ISO 27001Aligned
  • CycloneDX 1.6CBOM export
  • TLS 1.3PQ-only
Why this is urgent

The data you encrypt today is the data they decrypt tomorrow.

Adversaries don't need a quantum computer to start hurting you. They need a packet capture and patience.

Harvest now, decrypt later

Encrypted business data traversing classical TLS today (customer PII, financial records, IP, healthcare records, source code, internal communications) is being captured and stored by nation-state and criminal actors who plan to decrypt it once Shor's algorithm becomes practical at scale. Every TLS 1.3 session you serve without ML-KEM is a gift wrapped for that future.

Regulatory retention has a multi-decade horizon

Most enterprise data isn't ephemeral. SOX retention is 7 years; GDPR's right-to-be-forgotten doesn't unwind a 2026 capture decrypted in 2032; HIPAA disclosure exposure runs 70+ years; medical, financial, and contractual records compound over a person's working life. The blast radius of a 'someday' decryption is measured in regulatory cycles, not quarters.

The NIST timeline is the enforcement timeline

NIST has formally deprecated the classical-only stack starting in 2030, with full disallowance by 2035. Enterprise procurement cycles are already gating renewals on a documented PQ migration plan. 'We'll get to it' has stopped being an accepted answer.

Capabilities

Four pillars, hardened end-to-end.

We don't ship a kitchen sink. Each capability below is built for a single job a regulated workload depends on, and verified end-to-end in our own continuous smoke suite.

Hybrid PQ-TLS termination

ML-KEM-768 + X25519 hybrid key exchange terminated in a hermetic Rust gateway. TLS 1.3 only, no classical-only fallback. AEAD selection limited to AES-256-GCM and ChaCha20-Poly1305. The PQ handshake terminates at our edge; the upstream hop to your origin is re-encrypted over classical TLS 1.3. Plaintext upstream forwarding is strictly reserved for Enterprise Azure Private Link deployments, where the gateway and your origin share a private VNet.

Zero-touch CNAME onboarding

Customers add one TXT record (proof of ownership) and one CNAME (proof of routing). Our pre-flight resolver verifies both gates before issuing certs; no broken validations, no stuck ACME orders, no silent failures. The dashboard surfaces exactly which gate is open and which is still pending.

CycloneDX CBOM export

Every cryptographic surface in the gateway (KEM, AEAD, signatures, hash, RNG) surfaces as a single CycloneDX 1.6 cryptography bill of materials. Auditors get a machine-readable artifact instead of a slide deck; security teams diff CBOMs across releases to track exactly when a primitive changed.

Post-quantum AI inference gateway

One OpenAI-compatible and Anthropic-native API in front of five upstreams: OpenAI, Anthropic, Google Gemini (AI Studio and Vertex AI), and AWS Bedrock. Every route is streamed token-by-token over the same hybrid PQ-TLS edge. Bring your own provider key: it stays inside a KMS boundary, and the invoice stays with your provider. Every call is quota-governed and audit-logged, and a dropped connection still settles its quota and writes its audit row.

AI gateway

Governed AI inference, built for regulated teams.

One OpenAI- and Anthropic-compatible API in front of five model providers, on your own keys. Routing and caching are table stakes; what your auditors will care about is the evidence layer that rides every request.

Audit trails you can prove, not just print

Every inference writes a tamper-evident record: metadata hashed, signed with Ed25519, and anchored into a Merkle root. An auditor can verify the chain independently of us. Plain request logs are an enterprise upsell elsewhere; cryptographically anchored ones are simply not sold.

PHI and PII stopped at the gateway

Prompts are scanned and sensitive fields redacted before a request leaves your tenancy, with per-tenant token quotas enforcing spend and exposure budgets. Built for healthcare and finance traffic, not bolted on after.

Failover that respects the stream

Provider outages are absorbed mid-flight: failover engages before the first streamed token, exact-match responses serve from cache, and a per-provider circuit breaker stops cascades with weighted load distribution across your chain.

Your keys, your providers, quantum-safe transport

Bring your own keys to OpenAI, Anthropic, Gemini, AWS Bedrock, and GCP Vertex. Keys live in a vault, injected per request, never in our logs. Every hop terminates on hybrid post-quantum TLS (X25519 + ML-KEM-768), so the transport is evidence too.

How it works

From classical to post-quantum in three changes.

No backend rewrite. No new SDK. No protocol fork. Two DNS records and a tenant configuration are the whole integration.

  1. Prove ownership

    Add a single TXT record at `_scrutari-challenge.<your-host>` with the token the dashboard shows you. The gateway's resolver picks it up within seconds and flips your tenant's verification gate green.

  2. Route traffic to the gateway

    Set a CNAME from your customer-facing host to `<tenant>.edge.scrutari.ai`. Our pre-flight resolver confirms the CNAME chain actually terminates at our edge before issuing certs; no broken validations, no silent failures, no rate-limit burns.

  3. Serve post-quantum TLS

    The gateway terminates the ML-KEM-768 + X25519 hybrid handshake on your hostname, then re-encrypts the upstream hop to your WAF or load balancer over classical TLS 1.3 (X25519, AES-256-GCM or ChaCha20-Poly1305). Plaintext upstream forwarding is reserved for Enterprise Azure Private Link deployments, where the gateway and your origin share a private VNet. Service code unchanged; browsers negotiate the strongest cipher suite they support.

Move your TLS posture to post-quantum before the timeline moves you.

Scrutari is in early access for SaaS teams whose largest customers require a documented PQ-TLS migration plan and who don't intend to rebuild their TLS stack to produce one. If that's your constraint, get in touch.