/* Variation B — Technical / Architecture-forward The product IS a network proxy. Lead with the proxy diagram as the hero. Lighter, paper-toned, more system-y. Reads like a developer infrastructure page (Stripe, Vercel, Resend) — but in the Registrum palette. */ const VariationBBrand = () => ( R/001 RegistrumAI ); const VariationBNav = () => ( ); const ArchDiagram = () => (
{/* Agents column */}
Your agent fleet
billing-reconciler
support-triage
code-reviewer
unknown-agent-04
{/* Proxy center */}
Network-layer proxy
R/001 RegistrumAI
  • Identify
  • Meter
  • Govern
  • Audit
<5ms overhead OpenAI- compatible
{/* Providers column */}
Every LLM provider
OpenAI
Anthropic
Google · Gemini
Groq
Mistral
); const VariationBHero = () => (
Enterprise AI agent governance

One endpoint.
Every agent.
Every provider.

A drop-in OpenAI-compatible proxy that meters, governs, and audits every LLM call across OpenAI, Anthropic, Google, Groq, and Mistral. Network-layer. Zero instrumentation. The agent code doesn't change.

Read the docs Join launch list
< 5ms
Proxy overhead
0
SDK changes
5+
Providers covered
100%
Of LLM traffic captured
); const VariationBProblem = () => (
The problem

Enterprises are deploying AI agents faster than they can govern them.

Engineering teams, vendor integrations, contractors, and shadow IT are all spinning up agents that call expensive, powerful AI APIs — without procurement approval, cost visibility, security review, or audit trail.

CASB tools can block api.openai.com entirely, but can't allow gpt-4o-mini while denying gpt-4o. Hyperscaler cost dashboards see only their own provider. Developer SDKs only capture what someone remembered to instrument.

RegistrumAI is purpose-built for the gap — at the network layer, where every call is visible whether the developer cooperated or not.

); const VariationBHow = () => (
How it works

Three steps. One afternoon.

  1. 01

    Point your agents at the proxy.

    Change one base URL in your LLM SDK. No code modifications, no library install, no developer mobilization. Any OpenAI-compatible client works on day one.

    {`base_url="https://gw.registrum.ai/v1"`}
  2. 02

    Define identity, policy, and pricing.

    Agents are fingerprinted from their API key. Apply model allowlists, data residency rules, DLP, and budget caps. Load your contracted rates into the pricing catalog.

    {`policy: { models: ["gpt-4o-mini"], regions: ["us-east"] }`}
  3. 03

    Watch the ledger fill in real time.

    Every call: agent, model, tokens, cost, disposition. Stream to your SIEM. Export evidence packs for SOC 2, ISO 27001, HIPAA, and the EU AI Act. Hand the report to your auditor.

    {`GET /v1/audit?since=24h → 1,284,011 records`}
); const VariationBFeatures = () => { const rows = [ { tag: 'Discovery & inventory', t: 'See every agent the moment it makes its first call.', bullets: [ 'API-key fingerprinting names agents without tagging', 'Shadow AI flagged from first request — no quarterly audit', 'Lifecycle: provision · monitor · suspend · deprovision', 'Inventory exports auditors and regulators ask for', ], }, { tag: 'Cost metering & budgets', t: 'Real-time spend, attributed to a cost center, at your contracted rates.', bullets: [ 'Customer-owned pricing catalog (your negotiated rates)', 'Per-agent and per-department budget caps, enforced at proxy', 'Forecasting and anomaly detection on consumption patterns', 'Chargeback reports in Finance-compatible CSV', ], }, { tag: 'Policy engine', t: 'Enforcement, not observation. Decisions before the call leaves.', bullets: [ 'Model allow/blocklists per agent or department', 'Data residency routing — EU stays EU, HIPAA stays BAA', 'DLP for PII, secrets, source code, prompt injection', 'Rate limits, time-of-day controls, approval workflows', ], }, { tag: 'Audit & compliance', t: 'A tamper-evident ledger of every call your fleet has ever made.', bullets: [ 'Hash-chained, append-only audit trail', 'Optional full prompt / response capture with PII masking', 'Evidence packs: SOC 2, ISO 27001, HIPAA, GDPR, EU AI Act', 'SIEM-native export (CEF/LEEF, Splunk, Sentinel, Datadog)', ], }, ]; return (
What it does

Four control surfaces, one proxy.

{rows.map((r, i) => (
{r.tag}

{r.t}

    {r.bullets.map((b, j) =>
  • {b}
  • )}
))}
); }; const VariationBFooter = () => ( ); function VariationB() { return (
); } window.VariationB = VariationB; window.ArchDiagram = ArchDiagram;