When software spends money on its own,
there's no way to undo a bad payment.
Stablecoin and AI-agent payments are irreversible — once sent, the money is gone. No chargeback, no dispute, no escrow. In July 2026 the industry standardized how agents pay: the x402 Foundation launched under the Linux Foundation with Visa, Mastercard, Stripe, Circle, Coinbase and Google at the table — and its charter says nothing about what happens when a payment goes wrong. Visa and Mastercard add chargebacks only when a card stays in the loop. Recourse is the recourse layer for the payments that are truly card-free — agent-to-agent, x402, on-chain, peer-to-peer. We don't move the money; we make it safe to move.
The trust layer, as building blocks.
Everything an autonomous agent needs to move money safely — exposed as a clean API and a 5-line SDK. Use one primitive or all four.
Escrow
Hold a payment in an audited smart contract instead of sending it. Conditional release on delivery, timeout, or dispute.
POST /v1/paymentsDisputes
When a deal goes bad, three AI judges — compliance, forensics, network graph — rule in seconds. Low-confidence cases escalate to a human arbiter who co-signs. eBay's 90%-automated playbook, for agents.
/dispute · /pof/validateThe Recourse Score
A Bayesian trust score for every counterparty, built from every dispute across the network. A credit score for autonomous commerce.
GET /v1/reputationPolicy & Underwriting
The score becomes an enforced action: block, escrow, or instant-guaranteed settlement — priced by risk, two-sided.
/v1/policy/evaluateBuilt for software that spends money on its own.
AI shopping & procurement agents
Agents buying goods, data, or services for a user or business — protected from a wrong or fraudulent purchase.
Agent marketplaces & platforms
Embed Recourse as your platform's trust layer — every agent on it inherits recourse, instantly.
API / data / compute sellers (x402)
Monetize agents via pay-per-call without fearing non-payment or fraud on irreversible micropayments.
AI-native finance, payroll & treasury
Higher-ticket autonomous payouts where a single bad transaction is expensive — gated, priced, and recoverable.
AI-agent companies call our API instead of paying directly.
Our customer is the developer building an autonomous agent — shopping, procurement, data/API buying. Their agent moves real money with no human watching. Five lines change that.
payVendor(vendor, 50_USDC) // irreversible. no recourse.
npm install @recourse/sdk const recourse = new Recourse({ apiKey }) // pay through escrow instead of sending money you can't get back const p = await recourse.pay({ to: vendor, amount: "50" }) await recourse.dispute(p.id) // bad delivery → refund const { score, decision } = await recourse.check(vendor, "50")
The demo below = this agent's experience
The run below is exactly what happens inside that developer's agent — simulated so you can watch it. In production a human never sees it; it runs thousands of times a day inside other companies' software. You are the invisible layer underneath.
We ship as an x402 facilitator AP2 / ACP compatible so we plug under existing rails — "keep your rail, add recourse."
Two ways to plug in.
In every case the customer is the team building the agent — they integrate once and pay the take rate; the end-user is the one protected.
1 · Direct — the SDK
npm install @recourse/sdk → wrap any agent payment in ~5 lines. The agent calls us directly: pay(), dispute(), check(). Fastest path, full control.
2 · Under the rails — x402-native, live today
Our API speaks x402 end-to-end: a spec-compliant 402 challenge on our own endpoints, facilitator-grade POST /v1/x402/verify (EIP-3009 signature verification), and POST /v1/x402/wrap — one call that adds escrow + dispute rights to any seller's x402 payment. Every payment verified through us mints a dispute-rights receipt wired into the dispute engine. "Keep your rail, add recourse" — minimal change.
On partnerships: we're compatible with the open standards the ecosystem is adopting (x402, AP2/ACP) — that's how we plug in without anyone's permission. Formal partnerships (e.g. a recommended-facilitator listing) are a go-to-market goal on the roadmap, not a current claim.
Watch an AI agent get scammed — and recover. Then watch the network learn.
Run a payment. The escrow holds, the agent disputes a bad delivery, and a panel of three AI judges rules in seconds — escalating to a human arbiter who co-signs when confidence is low. The agent is refunded, and the outcome updates The Recourse Score, which every other agent on the network now inherits.
The demo has its own console.
Run a live payment through the real API: escrow, an AI-agent dispute, a panel of three AI judges you can click into and interrogate, human escalation when confidence is low, and the score updating in real time.
▶ Open the Live DemoThe Recourse network — every agent's disputes build one shared trust graph (illustrative demo data)
| Vendor | Recourse Score | Deliveries | Lost disputes | Policy for next $50 payment |
|---|
🔄 Why a brand-new agent wins on day one — the moat.
A first-time agent has zero history of its own. On its own it would have to get scammed once to learn a vendor is bad. The moment it joins Recourse, it inherits the entire network's dispute history — it avoids losses it did nothing to earn the knowledge of. Every agent's bad experience protects every other agent. A competitor can copy our code in a weekend; they cannot copy this shared graph.
The Recourse Score is the moat.
Anyone can ship an escrow contract. The defensible asset is The Recourse Score — a credit score for autonomous commerce, built from every dispute outcome across the network (above). It only works if you're neutral across rails, which Stripe / Circle / Coinbase structurally can't be — they each own a rail. That's exactly why Visa, sitting between issuer and acquirer, is the trust layer — not the banks.
And the score is already a product with a price: the basic score is free (the funnel), while machines pay $0.01 per premium trust report — over x402, today — and the score's real power is that it's enforced: it gates every payment, sets the protection fee, and decides who qualifies for guaranteed instant settlement. See the live Score Network — 50 counterparties, updating in real time →
We don't charge for escrow. We charge for risk borne.
Escrow is a smart contract — near-free, so we price it that way. The fee scales with the loss we prevent, not the money we hold. And the loss is large: a bad payment is a 100% loss, a card chargeback costs a merchant up to 3.4× the transaction, and businesses lose ~$4.60 per $1 of fraud. Against that, 0.6–1.5% to make a payment recoverable is cheap — and it's below Escrow.com (0.9–3.25%) and card interchange (~1.8%).
Two-sided revenue, like a card network. The agent (payer) pays a protection fee for escrow + dispute + refund. The vendor (payee) pays an instant-settlement fee when they opt to be paid now and we front + guarantee it — factoring their receivable. Whoever gets the benefit pays.
Agent: protection
Escrow + dispute + refund. Scales with vendor risk (150 → 100 → 60 → 40 bps); cheapest when the vendor is trusted.
Vendor: instant settlement
Opt to be paid now instead of waiting out escrow — we front & guarantee it (~60 bps). Liquidity, so the vendor pays. Trusted vendors only.
Underwriting spread
In the instant tier we absorb the dispute risk — insurance economics, priced off the Recourse Score.
Comps: Escrow.com 0.89–3.25% · card interchange ~1.81% avg · chargeback all-in cost ~3.4× face · $3.75–$4.61 lost per $1 of fraud (2025).
From working demo to the default trust layer.
Next up: design partners + a third-party audit — the ~18–24 month window to own the trust-data flywheel before it consolidates.
Full roadmap, costs & use-of-funds in the deck.