Proof, not promises
Every AI call writes a signed audit record, hashed and anchored into a Merkle chain you export as an evidence pack and verify offline. Most gateways keep plain request logs behind an enterprise contract.
Ed25519 · Merkle anchored
Gateway live in production · Self-serve
The AI gateway for
regulated industries.
Route OpenAI, Anthropic, Gemini and Bedrock through one control plane, with per-tenant quotas, signed audit records, and post-quantum TLS on every hop. Two DNS records to deploy.
No contract · Cancel anytime · Bring your own provider keys
ML-KEM-768
FIPS 203 ready
Evidence pack
verify offline
Built for estates that cannot put a US SaaS proxy in the data path
Post-quantum TLS gatewayRust-native inferenceMemory-safe by designOn-prem & sovereign readyOffline-first architectureBYOK · keys stay yoursMerkle-anchored auditFIPS 203 + 204 readyPHI redaction at boundaryFCPA compliantTLS 1.3 onlyCycloneDX CBOM exportPost-quantum TLS gatewayRust-native inferenceMemory-safe by designOn-prem & sovereign readyOffline-first architectureBYOK · keys stay yoursMerkle-anchored auditFIPS 203 + 204 readyPHI redaction at boundaryFCPA compliantTLS 1.3 onlyCycloneDX CBOM export
Why Scrutari
Routing is table stakes.
Evidence is not.
Every gateway can fail over and cache. What regulated teams cannot buy anywhere else at a self-serve price is proof that stands up in an audit.
Quantum-safe you can cite
Hybrid X25519 + ML-KEM-768 on every hop, CycloneDX CBOM export, optional HSM-backed anchor signing. Post-quantum transport is becoming table stakes. Artifacts compliance can reference are the part that is not.
FIPS 203 + 204 ready
Flat-price governance
Five providers on your own keys, streaming failover, exact-match caching, PHI redaction and signed audit, from one flat monthly price. No per-model fees, no enterprise gate on the features regulated teams need on day one.
Growth from $99/mo
The flagship · Live in production
One Rust runtime.
Two surfaces you can buy today.
The AI inference layer runs on the post-quantum TLS edge. Same binary, same audit spine, same flat price.
Shipping nowMulti-provider
For buyers who cannot use Cloudflare.
One OpenAI-compatible, Anthropic-native API in front of five providers. Bring your own keys. The model invoice stays with your provider, the governance stays with you.
- Route by provider, model or tenant: OpenAI, Anthropic, Gemini, Vertex, Bedrock in one control plane
- Fail-closed token quotas: per-tenant, calendar-month UTC, enforced at the proxy. No surprise invoices
- Signed audit row per request: Merkle-anchored, retention you control, queryable from your SIEM
- PHI redaction at the boundary: before any model sees the payload
- On-prem, sovereign or SaaS: the posture LLM gateway startups cannot ship
Drop-in migration
# one line. keys stay yours.
base_url = "https://api.edge.scrutari.ai/v1"
# everything else unchanged
client.chat.completions.create(
model="claude-sonnet-4",
stream=True,
)AI gateway specs
Providers5 · BYOK
StreamingToken SSE
QuotaPer-tenant / mo
EnforceFail-closed 429
AuditSigned · Merkle
DeployOn-prem · SaaS
Live · self-serviceTLS 1.3 only
Ship post-quantum in two DNS records.
A hardened Rust gateway that terminates ML-KEM-768 + X25519 hybrid TLS in front of the backend you already run. No application changes, no second TLS terminator to replace.
- Hybrid handshake at the edge: X25519MLKEM768 by default, no configuration flag
- Upstream re-encrypted over classical TLS 1.3 to your WAF or load balancer
- No classical-only fallback: TLS 1.3 only, AEAD ciphers only
- CycloneDX 1.6 CBOM export for every primitive in the data path
- Hermetic Rust runtime: single-process attack surface, no GC pauses
Onboarding
# 1. point the record
api.acme.com CNAME edge.scrutari.ai
# 2. prove ownership
_scrutari TXT "v=sc1 t=a91f…"
✓ hybrid PQ TLS activeGateway specs
KEMML-KEM-768
ClassicalX25519
AEADAES-256-GCM
TLS1.3 only
FIPS203 + 204 ready
Deploy2 DNS records
Edge AI · Pilot track
Inference where the network isn’t.
The second product line: Rust-native AI on physical edge hardware, governed by the same quota and audit spine as the gateway. Delivered through pilot engagement.
Pilot · TRL 4
Industrial safety
Scrutari Edge-Guard
A physical Edge AI appliance for continuous structural monitoring. Detects pre-fall indicators: corrosion, fastener failure, structural deflection, in real time at the point of inspection.
Uptime24/7
CloudNone
AlertsSigned JSON
StageTRL 4
Pilot
Enterprise data integrity
Project Helios
A hybrid sovereign platform for healthcare and financial services. Illuminates the black box of denied claims and regulatory compliance with Rust-native forensic auditing.
StreamingGB-scale
AuditVerifiable
ModeOffline-first
ResidencySovereign
Context Plane · Early access
Your agents remember things. Govern what they remember.
AI agents keep notes, memory, and context that outlive a session and cross privilege boundaries. Most teams cannot say what an agent knew when it acted. Governed agent memory is now live in the gateway in early access, so that question has a provable answer.
Live today
One key per agent
Give every agent in a fleet its own API key behind one gateway: spend, limits, and the signed audit record attribute per agent, across every model provider.
Early access
Governed agent memory
A memory-tool-compatible store with redaction before storage, provenance on every entry, access control at read time, and retention and deletion that actually reach agent state. Live in production, running our own outreach fleet today.
On the bench
Guards and evidence
Context compaction at the edge, denial-of-wallet guards, and signed decision traces answering what the agent knew when it acted.
Early access ships behind the same gateway; packaged with Growth and Enterprise when it lands.
Industries
Where an unprovable AI call is a liability
Healthcare
PHI redacted at the boundary before any model sees it. Per-tenant token budgets and signed records you hand an auditor as an evidence pack.
Financial services
Tamper-evident trails for every model call your desks make. Fail-closed spend quotas and failover that keeps trading-adjacent workloads answering.
Defense & public sector
Memory-safe Rust terminating hybrid PQ TLS ahead of the migration deadlines, with BYOK model access and private-network isolation.
Industrial edge
Inference at the edge for mining, energy and heavy-industry monitoring. Zero cloud dependency, same quota and audit spine.
Engineering posture
Memory-safe by construction.
No roadmap required.
CISA and the FBI name shipping new code in memory-unsafe languages a product-security bad practice for critical-infrastructure software, and ask vendors to publish a memory-safe roadmap. Ours is one line: the data plane is already 100% Rust, and the CBOM export lets you verify it rather than take our word.
Read the architecture0
GC pauses
Deterministic latency whether you’re firing a safety alert or streaming millions of records.
100%
Compile-time safety
Buffer overflows eliminated before code runs. The compiler enforces it, no runtime checks.
CBOM
Verifiable, not asserted
CycloneDX 1.6 export lists every cryptographic primitive in the data path, so the claim is something your auditor checks rather than believes.
Pricing
Flat price. No enterprise gate.
Signed audit and PQ TLS are on every plan, not held back for a contract call.
Full pricing, limits and billing FAQStarter
Free
Post-quantum TLS in front of the backend you already run. Up to 5 routes, one workspace per person, no card required.
- Hybrid PQ-TLS (X25519 + ML-KEM-768)
- Automatic ACME certificate lifecycle
- Per-tenant rate limiting + concurrency
- CycloneDX CBOM export · 99.9% SLO
Most popular
Growth
$99/mo
Adds the governed AI gateway. Up to 50 routes. First 30 days free, no card required.
- Everything in Starter, plus:
- AI gateway: OpenAI/Anthropic-compatible API
- BYOK: OpenAI, Anthropic, Gemini, Bedrock, Vertex
- Signed, Merkle-anchored AI audit + PHI redaction
Enterprise
Custom
Private Link, dedicated engineer, procurement-ready terms.
- Everything in Growth, plus:
- Per-tenant Azure Private Link (no public hop)
- FedRAMP-inheritable network posture
- Dedicated Customer Success Engineer
What is the Scrutari gateway, and how do I try it?
It sits in front of an existing backend and terminates ML-KEM-768 + X25519 hybrid TLS 1.3 on the public edge, then re-encrypts the upstream hop over classical TLS 1.3 to your WAF or load balancer. Onboarding is two DNS records. Starter is free, self-serve, no card required.
Why Rust instead of Python?
Python AI stacks are computationally heavy and exposed to runtime errors and memory leaks. Rust gives compile-time memory safety without garbage-collection pauses, so models run continuously on edge hardware without crashing. It is the same standard aerospace and automotive are adopting.
Do my provider keys ever leave my control?
No. Bring your own keys. They stay encrypted inside a key-management boundary, and the model invoice stays with your provider. We govern the traffic; we never resell the tokens.
How do you handle unstable connectivity?
Local-first architecture. Edge nodes operate 100% offline with embedded databases in replica mode; when connectivity returns, bidirectional sync updates the central systems. The pipeline never stops.
How is healthcare and financial data handled?
Sovereign-first. Processing happens locally or inside approved data-residency boundaries, never routed through third-party cloud APIs. The Rust pipeline streams and audits large datasets with mathematically verifiable output, so every decision carries a traceable trail.
Can I run this on-prem or air-gapped?
Yes. On-prem, sovereign cloud and air-gapped deployments are supported on Enterprise, with encrypted OTA model updates and automatic rollback so estates stay current without physical access.
Sizing your 2030 migration?
Drop us in front of your stack.
No application changes, no rewrites, no second TLS terminator to replace. Two DNS records and you’re terminating hybrid post-quantum TLS in production.
Talk to engineering
Not a sales engineer reading from a doc.
Scrutari is a hybrid post-quantum TLS gateway. We terminate X25519MLKEM768 on the public side and forward to whatever you run behind it. Reach out when you are sizing your 2030 migration or evaluating drop-in TLS termination for a regulated workload.
Hybrid X25519MLKEM768 by default. No configuration flag.
Drop-in via DNS CNAME. Your existing stack stays untouched.
Architecture, threat model, and load-test numbers on request.
Nashville, TN, USApartnerships@scrutari.ai
Talk to engineering
Pick what you're after below. The form routes to the engineering team that owns that surface, and we respond within one business day.