The AI Gateway That Pays for Itself
And it carries your EU AI Act evidence.
Route your AI traffic through Kertus: semantic caching and model routing cut 20–40% off provider bills, measured against a published methodology. Patient and customer data is masked in Frankfurt before any provider sees it. And every request writes tamper-evident audit evidence, so your December 2, 2027 record is accumulating from day one.
Building AI is Easy. Operating AI as a Business Is Hard.
Two problems compound as your AI usage grows: one drains your budget, the other accumulates regulatory exposure.
Pillar A
Your AI bill grows; your visibility doesn't.
Pillar B
Compliance is a runtime obligation, and the clock now reads December 2, 2027.
What most tools say
What enterprises actually need
Enterprises need explainability, not just enforcement. A blocked request without an explainable, recorded reason is useless in an audit. That gap is what Kertus closes: the reason, policy, alternative, and the cited legal basis and source on every decision.
One Gateway. Four Returns.
Kertus sits between your applications and your AI providers. Keep your infrastructure: every request comes back cheaper, safer, governed, and evidenced.
SAVE
Semantic cache and model routing cut 20–40% off provider bills, measured against a published methodology.
PROTECT
PII masking in Frankfurt before any provider sees the request. Zero prompt retention by default.
CONTROL
Policy enforcement, quotas, and per-customer metering and billing: every request governed in flight.
PROVE
Hash-chained, regulation-mapped audit records on every request: decision, reason, policy, and alternative. Export for auditors on demand.
Cost lands the decision. Evidence keeps it. One gateway, one integration, four returns.
How it works
Works with any provider
Plus any OpenAI-compatible API endpoint.
Keep your infrastructure. Kertus adds the savings, the protection, and the evidence, on every request.
Core Capabilities
Everything you need to cut provider costs, commercialize, govern, and evidence your AI APIs. We separate what ships today from what is on the roadmap. No vaporware above the line.
Semantic Response Cache
Semantically similar questions are served from cache; the provider is never paid twice for the same answer. Masked text only, encrypted, with TTL.
Intelligent Model Routing
Routine tasks are routed to cheaper models automatically, by explicit rules. Frontier models stay reserved for the work that needs them.
Savings Ledger & ROI Dashboard
Every saved cent is measured per the published 4-layer methodology and reconciled to raw usage records, net of provider-native discounts.
Usage Metering
Track requests, tokens, images, minutes, or custom units, per key, project, customer, and model.
AI Billing
Subscription, pay-per-use, and hybrid pricing for your own customers, with per-customer billing summaries.
Runtime Trust
Policy enforcement before requests reach providers: masking, budgets, quotas, oversight, and EU routing, all fail-closed.
GDPR / DSGVO
Runtime PII masking in Frankfurt before any provider sees the request, zero prompt retention by default, and DSAR deletion.
EU AI Act: Human Oversight & Logging
Human-oversight controls (notify / enforce / pause kill-switch) and Article 12-shaped tamper-evident logging, accumulating ahead of December 2, 2027.
DORA / NIS2 Evidence
Obligation-mapped evidence exports and an ICT incident report for financial and critical-infrastructure operators.
Tenant Isolation
Customer-level quotas, access, and billing separation, adversarially tested.
Audit Evidence Chain
Tamper-evident, hash-chained records of every decision: policy cited, confidence, and the alternative offered, with a customer-runnable verifier. Regulator-ready.
Explainable Decisions
Every allow, mask, route, or deny returns the reason, the policy reference, and the alternative, in the response and the evidence record.
Auditability
Immutable runtime decisions and billing evidence, exportable on demand and independently verifiable offline.
Bring Your Own Infrastructure
Keep your provider and cloud: OpenAI, Anthropic, Gemini, Azure, Bedrock, Mistral, or self-hosted. Change one base URL.
Zero Provider Knowledge · FULL mode
FULL visibility mode ships today: encrypted credential injection, token metering, EU-sovereign routing, and cost governance.
Context Window Management
Automatically trim conversation history beyond your configured token threshold, keeping the system prompt and the most recent turns verbatim.
Compliance Brain · Regulatory Watch
A hashed, versioned corpus of the EU AI Act, GDPR/DSGVO, DORA, NIS2, BDSG, TTDSG plus EDPB, BfDI, and ISO/IEC 42001 records, checked daily against the official sources. Changes open human-reviewed items; policy never changes silently.
Legal-Basis & Source Citations
Every allow, mask, route, or deny carries its cited legal basis and regulatory source, in the evidence record and the response, selected deterministically from a knowledge base pinned to the watched corpus.
HMAC Request Signing
Cryptographic proof that every request to your backend was authenticated, compliant, and quota-checked by Kertus: HMAC-SHA256 over the exact bytes, replay-bounded, secret encrypted at rest.
Prompt Compression
Conservative whitespace and duplicate removal before the request reaches your provider. The measured token delta is priced at your input rate and reported as savings layer L4.
Compliance Packs · HEALTHCARE + FINANCE_INSURANCE
Sector bundles: extra medical/financial PII recognizers, MDR/BaFin/EBA sources under the Regulatory Watch, and sector citations on decisions. Included on GROWTH and above; €199/pack/mo add-on on SAVER and AUDIT.
AI Shield · Runtime Security
Deterministic prompt-injection and content-safety detection (every finding names its rule) plus extraction and distillation traffic analytics. €299/mo add-on on every plan; free for design partners.
Zero Provider Knowledge · OPAQUE / BLACK_BOX
ENTERPRISE modes for fully private AI architectures: HMAC-signed forwarding to your own backend with no caching, no routing, no token parsing, and, for BLACK_BOX, no payload inspection or masking by design.
Compliance Studio
ENTERPRISE: your policies, DPAs, and works-council agreements as an encrypted, versioned per-tenant RAG. Human-activated deterministic rules (model allowlists, deny patterns, data classification, purpose limitation, per-customer scoping) whose denials cite your own clause, an obligation map with an auditor-ready gap report, and a self-serve review dashboard where your DPO holds the human gate.
On-Box Assist & Native PDF
Studio drafts candidate rules from your documents using a frozen open-weights model on your gateway's own server, the same box where your documents already sit encrypted. Nothing leaves, not even for the assistant; every proposal stays human-activated. Native PDF ingestion included.
AI Shield · Learned Detection
A trained classifier flags paraphrased injection attempts the deterministic rules miss, and auto-containment clamps the rate of keys showing extraction signatures. The learned layer only ever flags; the deterministic layer keeps the last word on blocking.
Dedicated & Self-Hosted Deployment
The full gateway, embedder, and on-box assistant inside your own perimeter: your cloud, your datacenter, or fully air-gapped with human-gated corpus bundles. Documented self-install with a published egress allowlist; accepted when the doctor check suite is green on your hardware.
Self-Serve Checkout & Billing Portal
Stripe self-serve subscriptions and a customer-facing billing portal. Today: payment links and concierge invoicing from the monthly statement.
Roadmap items are shown for transparency and are not billed, not counted in measured savings, and not represented as available. Design partners shape what ships next.
The Compliance Brain
Today, every allow, mask, route, or deny returns its reason, policy reference, alternative, and the cited legal basis and source, recorded as tamper-evident evidence and selected from a knowledge base pinned to the watched regulatory corpus (EU AI Act, GDPR, DORA, NIS2, BDSG, TTDSG, EDPB, BfDI, ISO 42001). On ENTERPRISE, Compliance Studio adds your own policies: uploaded documents become searchable, and human-activated rules deny with a citation of your own clause.
WITHOUT KERTUS
Decision: DENY
// That's it. No reason. No source.
// Good luck explaining this to legal.WITH THE COMPLIANCE BRAIN · TARGET
Decision: DENY
Risk: High-risk AI system
EU AI Act Annex III · Employment
Required controls:
✓ Human oversight mandated (Art. 14)
✓ Transparency notice required (Art. 13)
✓ Bias monitoring required (Art. 9)
✓ Logging and auditability (Art. 12)
Confidence: HIGH
Sources:
· EU AI Act Art. 6 + Annex III §4
· GDPR Art. 22: automated decisions
· BfDI Guidance 2024 §3.2
· Tenant policy: AI-GOV-007 §1.4This is what enterprise legal teams, works councils, and regulators will want on every decision. Today Kertus already stores the reason, policy reference, and alternative as tamper-evident evidence; the Compliance Brain will add the cited legal basis and source shown above.
Regulatory Corpus
EU AI Act, GDPR/DSGVO, DORA, NIS2, EDPB opinions, ISO 42001, BfDI guidance, and sector rules, maintained, versioned, and retrieval-ready.
Private Per-Tenant RAG
Upload your DPAs, works-council agreements, internal AI policies, and vendor allowlists. Auto-indexed into your isolated compliance tenant.
Proprietary Compliance Ontology
A structured map of use cases → risk categories → obligations → required controls → evidence. The moat competitors cannot quickly copy.
Your AI Architecture Is Your Business
Kertus ships in FULL visibility mode on every plan, the deepest cost governance. The OPAQUE and BLACK_BOX modes on ENTERPRISE extend the same enforcement to fully private AI architectures.
Full Visibility
You use OpenAI, Anthropic, or Gemini directly. Store your provider key in Kertus and unlock token-level metering, EU sovereign routing, and provider cost governance: semantic caching and intelligent model routing.
ROI: Customers typically recover their Kertus AI subscription cost from provider token savings within the first billing cycle.
Opaque Provider
You have your own AI backend and won't expose which model you use. Kertus proxies to your backend. Meter on your own unit: per diagnosis, per page, per query.
(ENTERPRISE-exclusive)
Black Box
You expose nothing. Hospital. Government. Financial institution. Kertus sits in front of your system. Full compliance enforcement. Zero provider disclosure required.
(ENTERPRISE-exclusive. Provider identity: never stored or required)
What you always get, regardless of visibility mode
Three modes. One API. Same endpoint.
# Direct OpenAI integration
POST /v1/proxy/legal-analysis
Authorization: Bearer kai_live_abc
{
"model": "claude-sonnet-4-5",
"messages": [...]
}
# Response headers:
X-Kertus-Decision: ALLOW
X-Kertus-Provider: anthropic
X-Kertus-Cost: 0.12
X-Kertus-Unit-Source: PROVIDER_RESPONSE# Proprietary AI backend
POST /v1/proxy/document-analysis
Authorization: Bearer kai_live_xyz
X-Customer-Token: customer-secret
{
"documentId": "doc_8821",
"type": "contract-review"
}
# Response headers:
X-Kertus-Decision: ALLOW
X-Kertus-Cost: 0.25
X-Kertus-Unit-Source: UPSTREAM_HEADER
# (provider: never known to Kertus)# Hospital AI · zero disclosure
POST /v1/proxy/clinical-support
Authorization: Bearer kai_live_hospital
X-Kertus-Use-Case: medical-diagnosis
X-Kertus-Data-Classification: PERSONAL
# Response (ALLOW):
X-Kertus-Decision: ALLOW
X-Kertus-Cost: 0.50
X-Kertus-Unit-Source: REQUEST_COUNT
# Same endpoint, SPECIAL_CATEGORY → DENY:
X-Kertus-Decision: DENY
X-Kertus-Cost: 0.00Black Box (ENTERPRISE): cost appears on ALLOW decisions, zero on DENY, because denied requests are never forwarded and never billed. Compliance enforcement without provider knowledge, and without payload inspection by design.
From AI Prototype to Commercial Product
Five steps to launch your AI API with cost governance and enterprise-grade trust.
Register AI Service
Define your AI service endpoints and configuration in the Kertus dashboard.
Configure Pricing, Quotas & Policies
Set up subscription tiers, usage-based pricing, and customer quotas. Configure caching, routing, and masking policies for your workload.
Connect Provider Endpoint
Point Kertus to your existing AI infrastructure: OpenAI, Anthropic, or self-hosted.
Route Requests Through Kertus
Keep your provider SDK: change one base URL to route through the Kertus gateway.
Get Savings, Enforcement & Evidence Instantly
Every request is classified, enforced, cached or routed for cost, masked, metered, and logged with an explainable decision (the reason, policy, and alternative), automatically.
POST /proxy/v1/document-ai
Authorization: Bearer customer-key
Content-Type: application/jsonSimple integration: just change the endpoint URL.
// Compliance decision header returned on every request
X-Kertus-Decision: DENY
X-Kertus-Reason: GDPR Art. 9: special-category data
X-Kertus-Policy: AI-POL-004 §2.1
X-Kertus-Risk: Cross-border transfer: provider not approved
X-Kertus-Alternative: Route to mistral-eu with PII masking
X-Kertus-Confidence: HIGH
X-Kertus-Sources: euaiact://art-6, tenant://dpa#4.3This is what closes enterprise deals. Not just "blocked", but why, with sources, and what to do instead.
Privacy by Default
Kertus does not store prompts or AI responses by default.
We only store the metadata needed for billing and compliance:
Your customer data flows through; it never stays with us.
Four Adjacent Categories. One Intersection.
Each neighbouring category covers one piece of the problem. Kertus sits where they meet.
AI gateways
Portkey, Helicone, LiteLLM
Routing and caching, but no regulation-shaped evidence and no customer billing
GRC platforms
Credo AI, OneTrust, Kertos
Documentation outside the request path, so it cannot produce per-request evidence
AI security suites
Check Point, F5, Palo Alto
Enterprise firewalls; no cost governance, no audit chain, built for 2,000+ employee companies
%-of-spend governance vendors
Usage-priced AI governance
Their revenue grows when your AI bill grows. Ours doesn't: flat subscription, savings measured.
Kertus AI
The intersection
Runtime enforcement + audit evidence + measured cost savings + EU sovereignty and customer billing, in one gateway.
Deep Dive
The Runtime Bouncer for AI Compliance
13 minutes. Two AI hosts. Everything a CTO, CISO, or compliance officer needs to understand why runtime enforcement is the missing layer in every EU AI Act strategy, and what Kertus AI does about it.
Recorded May 2026, before the EU Digital Omnibus agreement. The high-risk deadline discussed is now December 2, 2027; the runtime-enforcement argument is unchanged.
The Runtime Bouncer for AI Compliance
NotebookLM AI Podcast · 13 min 51 sec
Topics covered
AI-generated podcast. No cookies. No tracking.
Who It's For
Built for teams where the AI bill and AI governance both land on someone's desk. Kertus answers to both.
Engineering & CTO at AI scale-ups
Your token bill grows faster than revenue. Semantic caching and model routing cut it, measured on your dashboard and reconciled to raw usage records.
DPOs in healthcare
GDPR Art. 9 special-category data must not reach a provider unmasked. PII is redacted in Frankfurt before any provider sees the request, with evidence per request.
CISO & COO in fintech and insurance
DORA evidence obligations apply today; Annex III high-risk obligations land December 2, 2027. The audit chain accumulates from your first routed request.
AI SaaS founders
Meter, quota, and bill your own customers without building the infrastructure. Per-customer usage, budgets, and invoice-ready summaries out of the box.
Built to Scale With You
Start on SAVER for cost governance and compliance evidence from day one. AUDIT covers enforcement-only deployments. Fixed monthly pricing, no consumption surprises.
ROI guarantee: if your measured savings are less than your subscription fees over the first 60 days, we credit the difference.
SAVER and above. Requires ≥500 requests/month through the gateway. One claim per tenant. Credit, not refund. Savings measured per our published methodology.
Design Partner Program
5 of 5 slots remaining5 slots. €99/month locked for 24 months. White-glove onboarding, direct roadmap input. In exchange: a real production workload, a quotable savings metric, and a case study. All Compliance Packs included at no charge.
All plans are 100% SaaS: build once, sell many, customise via product. No forks, no bespoke engineering per client. Start on AUDIT or SAVER and upgrade at any time.
AI Shield (runtime security layer): €299/mo add-on on every plan; included free for design partners.
What counts as a service, and what as a customer?
Dedicated & Self-Hosted Deployment
The full gateway, embedder, and on-box assistant running inside your own perimeter: your cloud account, your datacenter, or fully air-gapped. For institutions that cannot accept any shared operator. Acceptance is mechanical, not contractual prose: the engagement is done when the doctor check suite is green and the evidence verifier passes on your hardware.
What you are paying for
- The watched regulatory corpus and citation knowledge base stay alive: daily source checks (or signed corpus bundles with every release when air-gapped), human-reviewed changes, and decisions that keep citing current law. A frozen gateway cites stale law; that is why this is a license, not a one-time purchase.
- Signed releases at least monthly: security patches, new named promise tests, and a changelog your auditor can actually read.
- A defined support allowance plus the acceptance tooling (doctor, the offline evidence verifier) your own team uses to prove the installation stays healthy, release after release.
- Flat by design: the gateway sends no telemetry home, so we could not meter your usage without breaking the zero-egress promise. Flat pricing is the honest consequence, not a marketing choice.
The Missing Operating Layer of the AI Economy
AI will not scale on prototypes and spreadsheets.
The future requires a compliance brain that reasons, enforces, and proves, not a static rule engine that says "denied" with no explanation.
Kertus is building that brain. Runtime enforcement, audit-grade evidence, and cited legal reasoning on every decision ship today. Built for the companies who cannot afford to get AI governance wrong.
Design Partner Application
Apply for a design partner slot, or just tell us about your AI workload. We respond within one business day.
How Kertus Works in the Real World
A realistic example of how companies integrate Kertus AI in minutes, without changing their product experience.
From AI Prototype to Commercial Product in Minutes
MedFlow Analytics
AI-powered clinical decision support platform
"The token bill doubled in a quarter, and getting legal and the works council to approve the AI took longer than building it."
Integrated Kertus in under 5 minutes
Before
fetch("https://api.medflow.ai/clinical-decision")After
fetch("https://proxy.kertus.ai/v1/clinical-decision-support")One endpoint change. Everything else stays the same.
Your Users Never Notice the Difference
Before Kertus
"Analyze Patient Risk"
Risk score
0.72
After Kertus
"Analyze Patient Risk"
Risk score
0.72
The customer experience stays exactly the same. Kertus works invisibly behind the scenes.
What Kertus Does Automatically
Request flow
Automated by Kertus
Cost governance + commercial controls + enterprise trust, automatically.
What a Real Request Looks Like
POST /v1/clinical-decision-support
Authorization: Bearer customer-key
{
"patientId": "anon-12345",
"analysisType": "risk-assessment",
"dataScope": "vitals-only"
}Kertus authenticates the customer, classifies risk, applies policy and masking, checks the semantic cache, routes for cost, meters usage, and forwards the request automatically.
Successful Request
// Headers
X-Kertus-Usage: 238/500
X-Kertus-Cost: 0.20
X-Kertus-Cache: MISS
X-Kertus-Request-Id: req_92384
X-Kertus-Policy: PASSED
X-Kertus-Privacy: VERIFIED
// Body
{
"riskScore": 0.72,
"confidence": 0.91
}The customer receives the exact same AI response, plus enterprise-grade usage visibility.
When Kertus Protects Your Business
429 Too Many Requests
{
"error": "QUOTA_EXCEEDED",
"message": "Monthly request quota exceeded."
}No provider call happens. No unnecessary AI cost is incurred.
403 Forbidden
{
"error": "SUBSCRIPTION_INACTIVE"
}Kertus blocks unauthorized usage before provider costs occur.
X-Kertus-Decision: DENY
X-Kertus-Reason: GDPR Art. 9: special-category data
X-Kertus-Policy: AI-POL-004 §2.1
X-Kertus-Risk: Cross-border transfer not approved
X-Kertus-Alternative: Route to mistral-eu + PII masking
X-Kertus-Sources: euaiact://art-6, tenant://dpa#4.3Not just blocked: explained, cited, and an alternative provided. This is what a regulator and works council actually want to see.
"The gateway cut the provider bill enough to cover itself, and we went from 'legal won't approve this' to 'here is the audit evidence, here are the sources' without rebuilding anything."
That is the gateway that pays for itself.
Frequently Asked Questions
Common questions about Kertus AI and how it works.