Skip to content
Live on Base Sepolia · 7 attestations seeded

AI's every decision,
on the record.

Drop in one SDK call. Every output your AI agent produces gets a cryptographic receipt — independently checkable by your customers, your auditor, your insurer, your regulator.

Trust topology · live
Watch a single decision earn its receipt.
streaminganchored
AI AgentClaude · GPT · …LLM Callprompt → responseTool Callsshopify · stripeDecisionapprove · rejecttrace.ai SDKone line of codeCanonical HashSHA-256 hashMerkle Batch→ one merkle rootBase L2public blockchainPublic Verifieranyone, anywhereAGENTTRACE.AIPUBLIC LEDGER
7
On-chain attestations
6
Verification axes
< $0.01
Anchor cost
2
First-party wrappers
The 30-CEO signal
87%of enterprise CEOs we interviewed name AI output verification as their #1 deployment bottleneck.

“The model is not the bottleneck anymore.
Trusting what it just did is.”

SourceShoptalk 2026 · Las Vegas — 30+ on-record CEO interviews.
Regulatory clock
Republic of Korea
AI Basic Act
LIVE
In force · day +137· January 22, 2026

High-impact AI providers must keep auditable records of training data, decision logic, and user-facing outputs.

European Union
EU AI Act · high-risk obligations
COUNTDOWN
T-minus 55 days· August 2, 2026

Article 12 requires automatic event logging across the AI system's lifecycle — traceable, tamper-resistant, retained 6+ months.

The receipt your auditor will ask for in 18 months is the same receipt we cut on day one.

The receipt question

You already log this. A log is not a receipt.

Your database keeps a record for you. A receipt is what you hand to someone who does not trust you. Five reasons the first cannot do the second.

Your database log today
An internal record
postgres · cloudwatch · datadog
  • Who can edit it
    Any admin with write access. Silent UPDATE, no audit on the audit.
  • How is it proven
    Trust the operator. There is no math an outsider can run.
  • Where it lives
    Behind your firewall. Disclosure requires legal, NDAs, and time.
  • Survives a breach
    No. The same intruder who falsified the decision can wipe the log.
  • Conflict of interest
    The party who acted is the party who keeps the record.
VerdictUseful for engineering. Not admissible for trust.
A trace.ai receipt
A public, signed record
DR-1 · merkle · base sepolia
  • Who can edit it
    Nobody. Once anchored, the bytes are sealed by a public chain — even we cannot rewrite them.
  • How is it proven
    Merkle proof + on-chain attestation. The math runs in the auditor’s browser, offline.
  • Where it lives
    Public Base L2 ledger. Anyone, anywhere, with a URL.
  • Survives a breach
    Yes. The hash is already on a chain you do not own — tampering only makes the proof fail louder.
  • Conflict of interest
    Separate notary wallet. The record is not held by either side of the dispute.
VerdictCryptographic. Independent. Survives the company.

We are not a replacement for your database. We are the layer that turns the line in your database into a receipt the outside world can verify without asking your permission.

How it works

Five steps from your AI to a public receipt. No new tools to learn.

  1. 1
    Step 1Listen

    When your AI agent acts, we listen.

    Wrap an Anthropic or OpenAI client, or create a DR-1 record at the decision boundary with the builder.

    SDK wrappers · DR-1 builder
  2. 2
    Step 2Format

    We give the decision a passport.

    A standard, language-neutral record auditors recognize — what the AI saw, what it considered, what it picked, and why.

    DR-1 schema · 12 fields
  3. 3
    Step 3Bundle

    Many decisions, one tamper-proof stamp.

    Hundreds of records get folded into a single fingerprint. Change one byte anywhere — the fingerprint breaks.

    Merkle batch
  4. 4
    Step 4Anchor

    Sealed with public-blockchain ink.

    The bundle stamp is written to a public, neutral chain. Once it lands, no one can erase it — not even us.

    Base L2 · EAS
  5. 5
    Step 5Prove

    Anyone can verify it. Forever.

    Customers, regulators, auditors, your insurer — all hit the same public URL. Six independent checks must pass.

    Public verifier · printable receipt
Positioning

What we are. What we are not.

Plaid for AI agents. Carfax for AI decisions. We sit between operators and verifiers — and never on either side.

we are

A neutral notary.

An independent attester anchors every Merkle root on a public chain. Operators can't rewrite their own logs.

we are

An integrator.

LLM client wrappers, Merkle trees, on-chain attestations. Battle-tested public goods, assembled into a new primitive.

we are

An open protocol.

DR-1 is MIT-licensed and proposed for ISO/IEC 24970. The standard is the moat — not the implementation.

we are not

Not insurance.

We don't underwrite risk. We don't pay claims. AI insurers are our customers, not our competitors.

we are not

Not custody.

We never hold customer funds. Ever. The platform attester wallet pays gas — that's the only money we touch.

we are not

Not a verdict.

We don't decide who's at fault. We make the facts cryptographically certain. Judges decide what they mean.

Pricing

Plain numbers. Both sides of the receipt.

Operators pay to anchor. Verifiers pay to check. Neither side subsidizes the other — that is what keeps us neutral.

Starter
$0forever

For developers wiring trace.ai into a prototype. Real attestations, on testnet.


  • 1,000 anchored decisions / mo
  • 1 AI agent
  • Base Sepolia (testnet)
  • Public verifier URL
  • DR-1 schema validation
  • Community support
Growth
$299per month

For teams shipping AI agents to customers. Mainnet, branding, support.


  • 100,000 anchored decisions / mo
  • Up to 10 agents
  • Base mainnet anchoring
  • Branded verifier domain
  • Print-ready receipts (PDF + QR)
  • 99.9% uptime SLA
  • Priority email support
Enterprise
$1,499starting at

For regulated industries: finance, health, legal. Built for auditors who arrive unannounced.


  • Unlimited decisions
  • Unlimited agents
  • Dedicated notary wallet
  • VPC / on-prem option
  • Custom DR-1 extensions
  • SOC 2 · ISO 27001 · GDPR
  • Insurance partner integration
  • 24/7 incident response
The other side of the market
OPERATOR SIDE

The team running the AI

Pays a flat subscription to anchor.

Predictable monthly cost, regardless of how often anyone audits you.

Why subscription: you anchor on a schedule — every decision, every day. Volume is yours to forecast.

Starter · Growth · Enterprise
VERIFIER SIDE

Auditors · insurers · regulators

$0.001 per proof verification — no minimum.

Pay only when you check. A regulator who never opens a case never pays a cent.

Why consumption: verifiers cannot lock into an annual contract with a notary they may one day need to challenge in court.

Pay-per-proof API

Either side can walk away. That is the point — a receipt nobody is forced to keep is a receipt no one is forced to trust.

Competitive landscape

Adjacent, not overlapping. Three categories. One open quadrant.

The market has tooling for the AI team, dashboards for the compliance team, and insurance for the CFO. Nobody is shipping receipts the outsider can verify without permission.

Feature matrix · who does what
CapabilityLangSmithCredo AIArmillatrace.ai
Cryptographic proof (Merkle + signature)···
Anchored on a public blockchain···
Independent verification by outsiders·
Survives company shutdown···
Open schema (DR-1, MIT)···
Developer-facing AI tracing··
Policy & governance frameworks···
Risk transfer (insurance payouts)···

The other three are real, growing, and necessary. They are also, taken together, what trace.ai customers buy alongside us. Receipts are the substrate the rest sits on.

Roadmap

From a single vertical, to the receipt of record.

Three phases. Each one builds the receipts the next phase rests on. No grand pivots; the schema we ship today is the schema regulators will reference in 2028.

012026
Live

MGA

Win one vertical decisively.

  • Bloom Co. live with anchored receipts on Base
  • KAIA & EU AI Act compliance kit shipped
  • 3 paid design partners across e-commerce, finance, healthcare
  • DR-1 schema v1.0 frozen, MIT-licensed
Shipping today
022027
Next

Platform

Open the receipts market to anyone.

  • Multi-tenant SDK · public verifier API
  • Third-party auditor & insurer integrations
  • $0.001/verification metered marketplace
  • Multi-chain anchor support (Base + L1)
032028+
Horizon

Standard

DR-1 becomes the receipt regulators ask for.

  • DR-1 referenced in jurisdictional guidance
  • ISO/IEC working-group alignment
  • 1B+ anchored receipts across operators
  • Receipts-as-collateral for AI insurance
North star

Phase 1 is what we sell today. Phase 2 is what we expand into with the receipts already anchored. Phase 3 is what regulators converge on once the schema is in production at scale.

Quickstart

One line. That's the integration.

Drop into any AI agent. We piggyback on the standard instrumentation tools you may already use; no new vocabulary to learn.

// pnpm add @vibingminers/sdk @anthropic-ai/sdk
import Anthropic from '@anthropic-ai/sdk';
import { traceClaude } from '@vibingminers/sdk';

// One wrap. Every messages.create now ships a receipt.
const claude = traceClaude(new Anthropic(), {
  agentId: 'cs-agent-v3',
});

const response = await claude.messages.create({
  model: 'claude-opus-4-7',
  messages: [{ role: 'user', content: prompt }],
  trace: { decisionClass: 'approve',
           rationale: 'within refund window' },
});

// → standard Anthropic response
// → receipt anchored on Base Sepolia in < 60s
// → verifier URL logged: /verify?id=<decision_id>