PAYMNTS — Assurance for the Agent Economy

CAPABILITIES

Trust, engineered in layers

Six independent systems, one verdict — so no single failure lets an agent run wild.

deterministic

Mandate verification

Every transaction must cite a live, signed mandate: amount cap, merchant scope, expiry. Violations hard-decline in strict mode — not "warned", declined.

5-factor fusion

Real-time risk scoring

Mandate compliance, sanctions proximity, behavioral anomaly, merchant risk, and velocity exposure — including cross-agent collusion patterns — fused into one score with evidence attached.

OFAC·UN·EU·FATF

Sanctions shield

Counterparties screened by name, alias, and graph proximity before approval.

per-agent memory

Behavioral baselines

Each agent builds a normal-behavior profile over time — anomalies surface as scored deviations, not guesses.

compliance

SAR autopilot

Suspicious Activity Reports generated from real evidence: facts, scores, graph paths, audit trail.

audit-ready

Provenance graph

Every decision carries its evidence chain — who mandated, what scored, why it cleared. Auditors query the trail directly.

agents + teams

Agent-first API

Agents verify mandates and check risk natively in their own flow; your team reviews escalations in the console.

strict mode

Hard-decline policy

Compliance overrides beat weighted scores — sanctions hits always stop the money.

always-on

Live health

Every check, verdict, and override streamed live — you see system health before your customers feel it.

trust anchors

Counterparty key registry

Merchant, trusted-agent, and issuer public keys governed in one registry — registered, rotated, and revoked under human-admin control with full audit. Unknown trust material escalates; it never silently verifies.

verifiable artifact

Issuer evidence bundle

Every scored transaction exports a signed evidence bundle — mandate, authorization, risk verdict, screening, provenance, and the hash-chained audit trail — verifiable offline and rendered as a document issuers review.

HOW IT WORKS

Three beats. Zero blind spots.

Agent registers → human signs

Agent gets a graph identity bound to an owner. The owner signs a mandate: amount cap, merchant scope, expiry — Ed25519-signed, tenant-isolated.

Agent attempts a transaction

Via ACP, Agent Pay, VIC, or AP2, the event normalizes into our pipeline — mandate check, sanctions screen, behavioral analysis, velocity check.

Verdict + permanent record

Approve, escalate, or decline — with scored evidence and a 7-year audit trail. Regulators get provenance, not a shrug.

WHO IT'S FOR

Every rail agents will spend on

Wherever an AI agent can initiate a payment, there's a compliance question someone will ask. These are the teams who'll hear it first.

Payment orchestration

PSP routers and checkout platforms score agent traffic before choosing a rail — clean agents take the fast path, risky ones get stepped up.

Orchestrators · gateways · checkout platforms

Card issuing & wallets

Agent-provisioned cards bound to a mandate — spend cap, merchant scope, and expiry enforced per-authorization, not just per program.

Card issuers · agent wallets · spend platforms

Banks & treasury

Agent-initiated sweeps, transfers, and reconciliation — every movement scoped to a signed mandate with an audit trail regulators can read.

Bank innovation teams · treasury · corporate banking

Custody & digital assets

Agents driving wallet operations get screened beyond policy scope — counterparty graph proximity catches what allowlists can't.

MPC custodians · exchanges · institutional wallets

Marketplaces & commerce

Agent buyers purchasing on behalf of users — mandate-verified orders with evidence that settles "who authorized this" disputes instantly.

Marketplaces · e-commerce · procurement agents

Fraud & identity platforms

Risk engines built for human telemetry get the missing agent signals — mandate compliance, behavioral deviation, counterparty context — as a feed.

Fraud vendors · identity platforms · risk consortia

SANDBOX ACCESS

Try it yourself — no sales call

The sandbox runs the full assurance engine on instant rails: request an account, mint an API key, and get a real approve / escalate / decline decision in under a minute.

Get an account

Request sandbox access (or register directly on self-serve deployments), then click the verification link we email you — verified email is required for keys.

Mint an API key

Badge in, open API Keys in the console, and mint a starter key — the secret is shown once, stored only as a hash, revocable instantly.

Score a payment

Call /risk/fast-check with your key — a real decision with evidence, typically under 200 ms. Declines never touch the rail; then watch it in the console's audit feed.

CONNECT

Three integration surfaces — one assurance engine

Same scoped, rate-limited assurance engine behind every door: REST for services, MCP for agents, OpenAPI for typed SDKs. Request a sandbox account, verify your email, mint an API key in the console, and score a payment in under a minute.

🔌 REST API

Drop it in any service

# decide before the rail sees it — fast-check
curl -X POST "https://paymnts.ai/api/v1/pymnts/risk/fast-check?tenant_id=YOUR_TENANT&project_id=YOUR_PROJECT" \
  -H "x-api-key: YOUR_KEY" \
  -H "content-type: application/json" \
  -d '{
    "tx_id": "tx_9f2", "agent_id": "checkout-bot",
    "mandate_id": "mnd_1",
    "amount": "129.99", "currency": "USD",
    "merchant_category": "retail", "protocol": "fednow"
  }'

Scoped by tenant + project. Every response carries X-RateLimit-* headers for your tier.

🤖 MCP server

Native tools for agents

// claude_desktop_config.json / any MCP client
{
  "mcpServers": {
    "paymnts": {
      "url": "https://YOUR-HOST:8121/mcp",
      "headers": { "x-api-key": "YOUR_KEY" }
    }
  }
}

verify_mandate · score_risk · check_sanctions · log_transaction · query_agent_history · check_merchant

📦 SDK

Typed clients, generated

# generate a typed SDK from the live spec
pip install openapi-python-client
openapi-python-client generate \
  --url https://YOUR-HOST/openapi.json

The spec is the contract — always in sync with the deployed routes, never hand-drifted.

BUILT FOR REGULATED TEAMS

Assurance, not custody

We never hold funds, never see your customers' payment credentials, and never participate in the rail ourselves — the sponsor bank executes our verdicts (release, hold, accept, reject); funds move only between your accounts there, never through a Paymnts account. Everything else is engineered to survive their audit.

🔐 HASH-ONLY KEYS

Secrets shown once

API keys are stored as SHA-256 hashes, TTL-enforced, and revoked instantly — the full secret exists only at mint time.

📜 TAMPER-EVIDENT AUDIT

Hash-chained, 7-year trail

Every decision, key issuance, override, and settlement lands in a SHA-256 hash-chained governance ledger, enforced at the database — tampering is detectable, not just prevented.

✍️ HMAC WEBHOOKS

Provider-signed events

Webhook deliveries verified by HMAC-SHA256 signature and made idempotent — retries never double-settle or double-decide.

🧱 TENANT ISOLATION

Scoped, hard-fails

Every query is tenant + project scoped — cross-tenant access returns 403, never silently falls back.

⛔ FAIL-CLOSED

Errors decline, never approve

If scoring or screening fails, the payment declines or escalates — there is no silent-approve path.

🛡️ HARDENED EDGE

TLS, HSTS, CSP, throttles

Secure headers on every response, per-key rate tiers, login lockout, TOTP MFA for console sign-in, and hash-only session tokens.

🔏 SIGNED MANDATES

Ed25519 verification

Every mandate is cryptographically bound to its owner — forged or replayed scopes decline at verification, not at the incident.

👥 FOUR-EYES RELEASES

Maker-checker dual control

High-value rail releases require a second admin's approval — a risk-approved transaction still can't clear on one signature alone.

📋 SOC 2 EVIDENCE

Control-mapped, on demand

A SOC 2 Security-TSC evidence bundle (CC6.1–CC8.1) generated from live data — the artifact your TPRM review or Type II auditor maps directly.

🤝 COLLUSION DETECTION

Beyond single agents

Agent rotation, shared-owner bursts, and merchant fan-in are scored as signals — one clean agent can't launder a ring's volume.

📦 SUPPLY CHAIN

Attested build

Every deployable ships a deterministic SBOM and dependency audit, anchored into the same governance ledger as your decisions.

📡 RESPONSIBLE DISCLOSURE

Researcher-safe

RFC 9116 security.txt and a safe-harbor disclosure page — report privately via /ui/security.

PRICING

Tiered for your traffic

Rate limits scale with your tier — every tier gets the full assurance engine, the difference is throughput and retention.

STARTER

Pilot

Core mandate verification + risk scoring
Sanctions screening (OFAC-format)
Community support

GROWTH

Production

Everything in Pilot, plus behavioral baselines
Protocol adapters (ACP · Agent Pay · VIC · AP2)
SAR generation + compliance exports

ENTERPRISE

Regulated

Everything in Production, plus custom limits
Dedicated infra + air-gap deployment
SOC 2 evidence bundle (CC6.1–CC8.1)
Four-eyes dual-control releases
SBOM + supply-chain attestation
Audit-grade retention + SLA