Now supporting Anthropic + OpenAI

Divine Shield against
data tampering.

Guardrails creates a tamper-proof, blockchain-anchored record of every AI call your application makes. One line of code. No infrastructure changes.

Start pilot — $2,500 / 60 daysRead the docs

Fee credited against your first annual contract.

Who needs Guardrails?

If your software talks to AI,
you need a record of what it said.

</>

Teams building with AI

Building a web app, a Python service, an Android or iOS app — anything that makes API calls to Claude, GPT, or another major AI provider? Guardrails drops in without touching your stack. Point your client at our proxy and every API call gets captured, encoded & cryptographically signed by two unique keys. The signed fingerprint is uploaded to our Verilink'ed secure blockchain, and the encoded prompt is stored on our secure servers for later review — acting as a release of liability for you, the customer.

WebPythonNode.jsAndroidiOSAnthropicOpenAI

Companies running AI in operations

If your business leans on AI for real decisions — triaging patients, reviewing contracts, approving claims, advising customers — you need permanent, unchangeable proof of exactly what the AI said and when. So if something goes wrong, you can show it wasn't on you.

HealthcareLegalFinanceInsuranceInternal toolsDecision support

What we do

Four steps from AI call to
irrefutable proof.

Guardrails sits invisibly between your app and your AI provider — sealing every interaction in a tamper-proof record.

YOUR APP1 line changeGUARDRAILSAI PROVIDERClaude / GPT
01 · INTERCEPT

Drop-in proxy

Point your existing Anthropic or OpenAI client at our Guardrails proxy URL and add three headers — your audit key, your provider key, and an actor ID. That's the entire integration. No SDKs to install, no code to rewrite. Every AI request now flows through us — invisibly, with under 50ms added latency.

SHA-256 · 0x3a7b…
02 · SEAL

Cryptographic fingerprint

We capture the full request, response, and reasoning trace, then create a SHA-256 fingerprint — a unique seal. It's dual-signed with your private key and ours. Change a single character later and the seal shatters.

ROOT500 events → 1 Merkle root
03 · ANCHOR

Blockchain-anchored

Every 30 seconds we batch 500 sealed records into a Merkle tree and commit a single root to Verilink'ed — our append-only blockchain. Each block locks the one before it. Tamper with history and the entire chain visibly breaks.

/verify/a3f8b21ctampered: falseblock #2020 confirmed
04 · VERIFY

Anyone can verify — no one can decode

Share a public proof link with a regulator. They see hash matched, signatures valid, block confirmed — no login, no trust required. The content of your AI calls stays encrypted and private. Only the proof is public.

“Guardrails is the Divine Shieldbetween your AI and your liability.”

Every framework. One gap.

HIPAA, EU AI Act, Joint Commission, Texas TRAIGA — they all point at the same requirement: an audit trail of AI decisions. None specify how to build it. Guardrails is the how.

FrameworkStatusRequirement
HIPAA Security RuleMandatoryAudit controls for PHI-touching AI
Joint Commission + CHAISep 2025Audit trail as 1 of 7 governance areas
EU AI ActAug 2026Transparency + human oversight
Texas TRAIGAJan 2026Document AI-assisted clinical decisions
California AB 3030In forcePatient disclosure for AI content

Start your 60-day pilot

$2,500 flat. Professional features, 100K calls, onboarding support. Fee credited against your first annual contract.

Get started