The payments rail
built for autonomous agents.

Curless is institutional payments infrastructure for the agent economy. One SDK for agent-native identity, scoped authorization, multi-rail settlement and tamper-evident audit — across cards, stablecoins and local digital wallets.

  agent_pay.ts — curgent-pay/sdk v0.7.4 · POST /v1/intents
1 // Authorize an agent purchase under a scoped mandate
2 import { Curless } from "@curless/agent-pay";
3
4 const intent = await curless.intents.create({
5   agent: "agt_chatgpt_5xQ2",
6   mandate: { cap: 42000, currency: "USD", scope: "travel.book" },
7   merchant: "mer_marriott_global",
8   rail: "auto", // card | stablecoin | local-wallet
9   audit: { kya: true, signature: "a-jwt" }
10 });
11
12 // → settlement.cleared in 380ms · chargeback-protected
Live settlement ledger Streaming
CG
ChatGPT · hotel.bookagt_cg_19a · MAR · 2026-05-08
$2,184.00 Settled
CL
Claude · saas.subscriptionagt_cl_88f · USDC · auto-renew
$ 79.00 Settled
GE
Gemini · b2b.procurementagt_ge_4d2 · ACH · pending
$ 12,400 Pending
GR
Grok · ads.bidagt_gr_71c · Stripe · 380ms
$ 0.04 Live
CR
Cursor · api.meteredagt_cr_5p9 · X402 · streaming
$ 1.62 Live
Macro thesis

The shift to agentic commerce is structural — and underway.

Internet traffic is no longer human-first. By 2030, autonomous agents will execute trillions in commerce annually — but the rails beneath them are fundamentally not agent-native.

$3–5T
Global agentic commerce GMV by 2030
McKinsey · 2026
51%
Of internet traffic is already automated
Cloudflare · 2025
43%
Of retailers piloting autonomous AI agents
BCG · 2026

“Existing infrastructure — identity, authorization, settlement and disputes — is fundamentally not agent-native. The window for first-mover infrastructure lock-in is 6–12 months.”

Problem

The agentic-commerce integration gap.

Four structural failures and a combinatorial standards problem make today's rails uneconomic, unsafe and unscalable for autonomous agents.

01 · Identity

No native identity

Agents are neither natural persons nor legal entities — existing KYC/KYB compliance frameworks do not map to a non-human transactor.

KYC/KYB does not apply
02 · Authorization

No native authorization

Legacy OTP/SMS flows require humans; agents cannot receive codes. Workarounds rely on virtual cards with hard caps — unsuited for recurring scoped mandates.

OTP / VCC does not scale
03 · Settlement

No native settlement economics

Card minimums and percentage fees render high-volume sub-$10 micropayments structurally uneconomic for autonomous agent transactions.

Micropayments uneconomic
04 · Disputes

No native dispute regime

Chargebacks assume a fraudulent “this wasn't me” scenario — the system fails completely when an explicitly authorized agent executed the transaction.

Chargebacks misfire
The N×M standards nightmare
In just 132 days, five incompatible agent-payment standards emerged — each requiring fragmented, breaking integrations.
M agents × N rails = M·N integrations
Agents
ChatGPT
Claude
Gemini
Grok
Cursor
M × N
3–6 eng-months / integration
Standards / Rails
Visa ICC
Stripe ACP
MC Token
X402
APOP
Solution

CurgentPay™ — one SDK for safe agent transactions on every rail.

Built-in identity, scoped authorization, policy and risk evaluation, and tamper-evident audit. Two pillars: accept payments from agents, and let agents pay safely on your behalf.

Pillar 01 · Receive Money In

Accept payments from agents — into your existing gateway.

Add agent-payment acceptance to existing PSPs and gateways with a single integration. Merchants accept inbound payments directly from ChatGPT, Claude, Gemini, Grok and customised enterprise agents.

1
Agent presents tenderAuthenticated request from LLM client.
2
Authorization gatewayIdentity (KYI), policy, and risk decisioning.
3
Merchant system of recordOrder placed via native commerce API.
4
Settlement & clearingCard, stablecoin, or local rail.
Use cases Hotels OTAs Marketplaces Merchants
Pillar 02 · Pay Money Out

Empower enterprise agents to procure — under mandate.

Enterprise agents execute secure, autonomous purchases through auto-renewing, mandate-verified spending limits — every action signed, logged and underwritten.

1
Principal issues mandatePre-authorized envelope (cap, scope, expiry).
2
Agent submits signed intentAP2 mandate (Intent → Cart → Pay).
3
Authorization gatewayIdentity, policy, risk and rail selection.
4
Settlement to merchantCard or stablecoin with cryptographic audit.
Use cases B2B Procurement Fintech Platforms

The Curless Guarantee

We strictly underwrite every agent transaction: payment-correctness guarantee, automatic dispute resolution, and chargeback protection. If your agent errors, we make you whole.

Architecture

A two-plane settlement engine.

Untrusted reasoning is strictly separated from trusted settlement. Agents issue signed payment intents; the core executes them under mandate, policy and tamper-evident audit.

Reasoning Plane · Untrusted Client Layer No key material

LLM-driven agents issue payment intents only.

No cryptographic key material. No direct payment-rail access. The reasoning plane proposes; the settlement plane disposes.

ChatGPT Claude Gemini Grok Cursor + Custom
Scoped Capability Token · A‑JWT
Settlement Plane · Trusted Core · CurgentPay™ Compliance Boundary

The trusted settlement core.

Identity, policy & risk evaluation, multi-rail routing, and tamper-evident audit. Built to the standards regulators apply to electronic money institutions, not merchants.

KYA / KYI
Identity for non-human transactors. Reputation graph.
Protocol Translation
Visa ICC · Stripe ACP · MC Token · X402 · APOP.
Policy & Risk
Mandate, velocity, geo, fraud and counterparty.
Multi-Rail Routing
Card · stablecoin · local digital wallets.
Inbound · Accept
Merchant accepts an authorized agent payment
  1. Agent presents tender. Authenticated request from LLM client.
  2. Authorization gateway. Identity (KYI), policy and risk decisioning.
  3. Merchant system of record. Order placed via native commerce API.
  4. Settlement & clearing. Card, stablecoin, or local rail.
Outbound · Procure
Authorized agent transacts on behalf of principal
  1. Principal issues mandate. Cap, scope and expiry envelope.
  2. Agent submits signed intent. AP2 mandate (Intent → Cart → Pay).
  3. Authorization gateway. Identity, policy, risk and rail selection.
  4. Settlement to merchant. Card or stablecoin · cryptographic audit.
Roadmap

Path to the Agent Bank.

Today, an integration layer. Tomorrow, an orchestration plane. At maturity, a balance-sheet bank for the agent economy.

Today
Phase 1

Standardization & Settlement Layer

Agent payment processing and standards translation. One SDK lets any agent safely receive and pay across all major rails — including cards, stablecoins and local digital wallets.

2027
Phase 2

Orchestration Layer

Agent-to-agent autonomous optimization. A native KYA engine, reputation graph, and policy guardrails enable real-time autonomous decisioning for best price, supplier and timing.

Today. A standardization & settlement layer Tomorrow. An orchestration layer At Maturity. A bank for the agent economy
Market opportunity

A $200B+ addressable fee pool.

Even a 2–5% blended take-rate across processing, compliance and AgentFi yields a fee pool larger than today's global card-acquiring market — concentrated in a new, agent-native infrastructure layer.

$3–5T
Global agentic commerce GMV by 2030
2–5%
Blended take-rate · processing, compliance & AgentFi
$200B+
Total annual Agentic Finance Expense at maturity

Sources: McKinsey “The Agent Commerce Decade” (2026); BCG payments fee-pool model; Curless analysis.

Annual revenue mix at maturity
AgentFi · credit + treasury$60–100B
Payment processing$50–100B
Compliance + identity$30–50B
Audit + dispute + data$10–20B
Business model

The agentic value flywheel.

More volume → richer data → sharper pricing and deeper credit → more volume. The same compounding loop that built every modern payments leader — built natively for agents.

1 · Processing
2 · Trust + Data
3 · AgentFi
4 · AgentCap
The Engine
compounding loop
Year 1
$1B
GTV under management
Gross Transactional Value (GTV) running through the standards & settlement layer.
Year 3 · Processing
$10B
Annual TPV
Total Payment Volume processed annually across rails.
Year 3 · Orchestration
$100M
Custodial AUM
Idle agent balances auto-swept into yield via AgentCap treasury.
Year 3 · AgentFi
$100M
Total credit disbursed
Subject-backed agent credit lines underwritten on track record.
Go-to-market

Enterprise-led market capture.

Three sequenced motions: design partnerships harden the core, marketplace distribution drives bottom-up adoption, and vertical dominance defends the moat.

2026 H2

Strategic Design Partnerships

Sign top-tier OTAs, marketplaces and fintech partners. Co-build use cases, harden the enterprise API, complete product-market-fit validation.

2026 H2 → 2027 Q1

AI Marketplace Distribution

Listed natively in OpenAI, Anthropic and Marketingforce marketplaces. Embedded into enterprise agent stacks for bottom-up mid-market adoption.

2027 Q2+

Vertical Dominance

Default choice for hotels, hospitality and online commerce. Horizontal expansion into high-frequency agents (smart-vehicle, B2B procurement).

Institutional traction & strategic partners
Sinomaster
SZ · 000034
Global Top-3 digital products distributor.
Marketingforce
HK · 02556
Leading marketing-AI platform for enterprises.
LianLian DigiTech
HK · 02598
China's leading cross-border payment provider.
Newland
SZ · 000997
Global payment terminals and IoT solutions.
Competitive landscape

Agent-native × enterprise-grade.

Stripe and Adyen retrofit agents onto a human-payments core. Banks haven't started. Curless ships the only fully agent-native, multi-rail, multi-region core today.

Capability Curless Stripe Adyen Traditional Bank
Agent-native architecture
Competitive pricing across all transaction sizes ~
Non-US (EU + APAC) compliance & data privacy
Multi-rail · card + stablecoin + local digital wallets ~ ~
AgentFi credit + AgentCap treasury ~
  Native, full coverage ~  Partial / plug-in retrofit   Not supported