About Pricing Open prototype Join waitlist

CadinalPay

Payment controls for AI-assisted purchasing.

CadinalPay is an agent-facing payment interface with a human-facing control room: agents get a programmable way to request purchases, while people see policy, approvals, receipts, and the reason every decision happened.

Join the waitlist Watch the agent demo Inspect guardrails See pricing

Live agent payment control

Buy one market report for this procurement decision. Max $50. No subscriptions.

Creating a bounded authorization request and comparing one-shot vendor routes.

Agent interface call

createPaymentIntent({
  amountCap: "$50.00 USD",
  purpose: "agent delegated purchase",
  policy: "no_recurring_billing",
  route: "sample-vendor.example/report-store",
  evidence: ["terms", "vendor", "finance"]
})

Payment control runAllowed

01Parse human intent

02Create authorization intent

03Evaluate policy

04Prepare evidence trace

CadinalPay verdictAllowed

Route inspectedsample-vendor.example/report-store

Human proofterms snapshot + tax receipt requirement + route quote

Open the agent demo

Why now

Agents are moving from conversation to action.

Agents are starting to buy reports, renew tools, and restock supplies. But every checkout was built for a human with a mouse, so an agent wanders multi-step web flows, burns tokens re-reading pages, and leaves finance no usable trail.

Built for people, not agents

Vendor checkouts assume a human will read, click, and judge each screen.

Open loops burn tokens

Driving a full website re-feeds page context into the model on every step.

No audit trail

Nothing records what was bought, why it was allowed, or what finance receives.

CadinalPay's answer

A structured payment interface: short, bounded calls instead of full-page browsing.

Product model

The UI is for humans. The interface is for agents.

A chat prompt is useful because it captures human intent. But the product is not a checkout chatbot. CadinalPay models the hidden machine interface an agent needs: create a payment intent, check policy, reserve context, and emit evidence.

The screen exists so a human can supervise that interface. It shows what the agent tried, which policy allowed or blocked it, and what evidence package finance receives.

System architecture

One request becomes a checked, bounded purchase.

Human intent translated

Turn a conversational request into a structured agent payment intent.

Agent contract called

Expose budget, merchant, purpose, policy context, and approval state to the agent.

Policy verdict returned

Allow, block, or escalate before a payment can be authorized.

Audit trail emitted

Attach receipt, tax, approval, and decision evidence to the transaction.

Human intent

Agent payment interfaceCreates a bounded payment intent the agent can call programmatically.

Policy engineReturns allow, block, or approval-required before authorization.

Evidence ledgerStores receipt, tax, approval, and jurisdictional decision evidence.

Payment rail

Parallel journey

The agent and the human move on one timeline.

CadinalPay keeps both sides in sync: as the agent works a purchase, the human sees the matching view and the controls to steer it.

StageAgent timelineHuman visibility & control

Goal receivedAgentParses the constraintsHumanSees the goal summary

SearchAgentBrowses and compares vendorsHumanLive feed: "Comparing 3 reports"

EvaluationAgentScores routes against policyHumanSees rationale and top pick

ApprovalAgentRequests sign-off if neededHumanApproves, edits, rejects, pauses

PaymentAgentExecutes through the protocolHumanReceives receipt and log

LearningAgentSaves the outcomeHumanAdds notes for next time

Agent demo

Use chat as the entry point, not the whole product.

The demo should feel partly conversational because humans naturally delegate tasks in language. The differentiator is the trace beside the chat: the agent's CadinalPay call, the policy verdict, and the resulting finance artifact.

Human promptNatural language delegation

"Get the report, stay under $50, do not start a subscription."

Agent-facing interfaceStructured payment intent

Merchant, cap, purpose, policy context, required evidence, and approval mode.

Policy responseAllow, block, or escalate

The agent receives an executable verdict instead of raw human judgment.

Human-readable proofDecision trace and receipt

The supervisor sees why a payment was allowed, blocked, or escalated.

Agent workflow

The one-report purchase.

A procurement analyst needs one market report to support a decision, not another annual subscription that finance has to unwind later.

Pre-flight overlay

Intent

"Buy the single market report needed for this procurement decision. Max spend $50. No subscriptions."

Search approved and public vendor index

Render vendor checkout because no API is available

Block if receipt lacks VAT ID or cancellation proof

Authorize one-time payment and save receipt metadata

Spend cap locked at $50

sample-vendor.example/report-store/checkout

Visual receipt captured

Screenshot, DOM snapshot, markdown extract, rail quote.

Session initiated

Reviewing sample vendor checkout

Extracting receipt metadata

Capturing evidence artifact

Feature - Budgetary throttle

Policy first, payment second.

Finance stewards need controls that read like policy, not API documentation. CadinalPay turns budget posture into visible, adjustable constraints.

Autonomous budget throttle

Daily agent budget$250.00

Unverified vendor limit$50.00

Conditional rules

If vendor is unverifiedand transaction > $50Require approval

If VAT evidence missingor auto-renew detectedBlock payment

If rail spread > 2.5%compare next routeRequote

Human-in-the-loop

The line between auto-approve and ask a human.

Spend size, vendor trust, and category decide how much autonomy the agent gets. Routine spend clears itself; anything risky stops for a person.

Under $50Routine, known categoryAuto-approve

$50 - $500Known, verified vendorNotify + log

Over $500Or any new vendorRequire approval

SensitiveRestricted categoryAlways ask a human

Repeated failureRetries exhaustedPause + escalate

Feature - Evidence ledger

Every ledger row carries the business context.

Payment speed only matters if the receipt is usable. CadinalPay keeps vendor, approval, tax, rail, and receipt evidence attached to the transaction.

Spread savings

$14.50

Sample data vendorCross-border

$48.00Card or account railTax receipt captured

Metered API vendorInternational

$62.00API purchaseRequoted

Trial subscription vendorInternational

$850/moHoldAuto-renew blocked

Render confidence dropped

Hidden iframe text: "automatic renewal after 3 days at $850/mo."

Spend stateFrozen before payment

Evidence preservedScreenshot + DOM diff

Operator actionReject recurring billing cycle

Edge case handoff

The agent fails visibly, then hands off cleanly.

If a checkout introduces subscription terms, missing receipts, or a spend limit breach, CadinalPay freezes the payment and sends the exact issue to a human.

Feature - Institutional memory

Every decision teaches the next purchase.

CadinalPay records what was bought and why it was approved or rejected. Agents reuse that shared log instead of re-deciding from scratch, so they adapt to your vendors, budgets, and approval style over time.

Decision logWhat was bought, and why

Each purchase keeps its reason, policy verdict, and the human who signed off.

Vendor trustLearned over time

Repeated approvals raise a vendor's standing; rejections lower it.

FeedbackTags and "never again" rules

A rejection with a note becomes a rule the agent applies next time.

EfficiencyShared logs, no repeated inference

The agent reads a prior decision instead of re-reasoning the whole task.

Business outcome

Designed to make the ROI legible in one glance.

$50sample single-purchase spend cap

3checks before payment authorization

1receipt package for finance review

0subscriptions allowed without approval

Indicative pricing

How CadinalPay is planned to be priced.

CadinalPay is pre-launch, so these plans are indicative - they show the intended shape of pricing, not a checkout. Join the waitlist and we'll confirm details before anything is billed.

Monthly

$299/ month

For teams testing controlled agent payments in one workflow.

1 agent workspace
$25K monthly authorized volume
Policy rules and approval handoff
Receipt and evidence ledger

Join the waitlist

YearlyBest value

$2,990/ year

For teams standardizing agent payment controls across recurring operations.

Everything in Monthly
Two months included
$350K yearly authorized volume
Priority onboarding and policy design

Join the waitlist

Pay as you go

1.5% of authorized spend

For low-volume pilots or teams that only want to pay when agents use CadinalPay.

No platform minimum
$0.30 per authorization event
Approval, block, and receipt events included
Upgrade to monthly anytime

Join the waitlist

Payment processor fees and pass-through merchant charges are not included. Enterprise caps, custom approval policy, and dedicated payment rails are quoted separately.

Waitlist

Get early access to CadinalPay.

CadinalPay is pre-launch. Join the waitlist if you run agents that need bounded payment authority, finance-ready evidence, and human escalation before risky spend - we'll reach out as we open access.

Join the waitlist Explore prototype