Credit that repays itself out of your sales.
A corner shop proves its business is real with a zero-knowledge proof, takes a small advance in stablecoins, and repays 8% of every sale until it is cleared. No collateral, no credit file, no name, no phone number.
The shops that move the most cash are the hardest to lend to.
A kiosco in Buenos Aires, a phone-accessory stall in Lagos, a duka in Nairobi. They turn over stock weekly and they are invisible to every underwriter, because the three things a lender asks for do not exist: collateral, a credit file, and audited books.
Identity is the wrong question
Lenders ask who you are. What actually predicts repayment is whether money moves through the till. We underwrite the shop, not the person.
Promises default; cash flows don't
Advances repaid out of a flow that turns over in days behave very differently from loans backed by a promise to pay later.
Privacy is not a feature here
A public ledger of who borrowed how much, shop by shop, is a safety problem in these markets. Proofs go onchain; credential details never do.
Four steps, and the repayment is automatic.
The merchant never fills in a form about themselves. Everything we need, they already have: a seller dashboard and a till.
Prove the business, not the person
With zkTLS the merchant proves their seller dashboard shows a country, a revenue band and 6+ months of activity. That becomes a Business Pass. We never ask for a name, a phone number or contacts. World ID first proves one unique human, not banned, with no open advance.
Get paid in stablecoins
Customers scan the shop's QR and pay in USDC. While an advance is active, 8% of each sale goes toward repayment. With no advance outstanding there is no holdback and no checkout fee.
Draw stock credit without re-approving every time
The merchant sets limits once — max per advance, max per month, recipient, expiry — and delegates with a World Agent Pass. A cash-flow agent then draws small advances to pay suppliers, and the contract enforces the mandate and tier caps onchain. Agents can only propose signed offers; the risk agent can only lower limits, never raise them.
Repay through sales, build a private record
Repayment comes out of the 8% split automatically. A fully repaid advance mints a Standing Pass: a Noir zero-knowledge credential that raises the merchant's limit without revealing any history. Good behaviour compounds; the record stays private.
What's onchain, and what never leaves the phone.
Every screen has a World view toggle that shows exactly this split, live. We would rather show the honest boundary than claim the whole thing is private.
Public, under a shop pseudonym
- Sales, repayments and advance sizes
- Every transaction — writes are public
- Proof verifications and AI decisions, decoded at
/proof
On the device only
- Seller-account identity and revenue detail — proven, never sent
- Business Pass and Standing Pass witnesses: proofs are built in the browser
- World ID, kept only as an HMAC on our server; nothing about the person goes onchain
- Browser reads route through Anon-RPC (Nox) by default. No analytics or trackers ship in the app.
Censorship resistance
If our API, relayer, issuer and agents all disappear, the contracts keep working. Anyone can call pay, repay, writeOff (for a 5 USDC bounty) and the vault deposits and withdrawals. Non-upgradeable, no pause.
Open source
MIT. The whole stack builds and runs locally with no real keys.
Privacy
The lead property, not a bolt-on: ZK passes, device-side proving, private reads by default.
Security
Tier caps, exposure cap, mandate limits and 20% junior cover enforced in the contract. Honest caveat: this is unaudited pilot code.
Cheap for the merchant, legible for the lender.
The fee accrues at 0.5% per week on original principal while principal remains, and caps at 3%. Paying sooner costs less. The 8% holdback sets the pace of repayment; it is not an extra fee.
| Time outstanding | Fee on $100 | Total repaid | vs a typical advance |
|---|---|---|---|
| 1 week | $0.50 | $100.50 | — |
| 2 weeks | $1.00 | $101.00 | — |
| 4 weeks | $2.00 | $102.00 | — |
| 6 weeks and beyond (cap) | $3.00 | $103.00 | factor-rate advances commonly charge 1.1–1.5× the amount advanced |
Fall behind the minimum (1% of owed per 7 days) and the fee doubles to 1% a week, capped at 6% in total — and 100% of that late fee goes to the reserve, the first-loss buffer that would absorb that merchant's own default. Late payers pre-fund their own risk; lenders are not paid extra for a worse book. On-time merchants see no change at all.
Where a fee goes
≈9.6% illustrative
At 75% utilisation, a 4-week cycle and 12 cycles a year, in a simple model. Not a forecast and not guaranteed.
First loss, traded
An ERC-4626 share that takes losses first and trades on a Uniswap v4 pool. At 0.25% loss per cycle the illustration is +21%; at 1% it is −24%. It can lose value.
$2,000 per $1M lent
Our 10% of fees at the four-week fee, before costs. Plus 25% of the RiskHook swap fee to treasury; 75% goes to the reserve.
The risk market
DRIP trades against USDC on a Uniswap v4 pool whose RiskHook does two jobs: it takes a USDC-side fee for the reserve, and it keeps a TWAP the risk agent reads as an early-warning signal. The honest limitation: the pool is team-seeded, so the price is a signal, not a validated credit oracle — which is why the agent can only lower limits, never below the floors, and why write-offs and repayment pace are the other two triggers.
This is running, not a mockup.
Built at ETHGlobal Mumbai 2026 and live on Sepolia since. What is real and what is still simulated, with nothing dressed up:
- LiveSepolia contracts: AdvanceBook, BusinessPassRegistry, senior vault and DripCoin ERC-4626 tranches, Uniswap v4 RiskHook. Non-upgradeable, no pause.
- LiveBrowser-generated Noir ZK proofs (
business_pass,standing_pass) verifying onchain, decoded at/proof. - LiveReal World IDKit and Agent Pass wired on the hosted app. AI underwriting plus a critic, managed cash-flow agents, risk agent, keeper and relayer running.
- LiveAnon-RPC private browser reads on by default — 26 of 26 reads routed in a Chrome measurement.
- Live875 offline tests: 169 Foundry, 554 agents, 152 SDK. Playwright across every route at phone and desktop sizes with accessibility checks.
- SimulatedSeller data: the demo uses a mock zkTLS attestor. The real Reclaim path is built and tested; it needs a working provider for a real dashboard (Stripe, Mercado Pago, Paystack).
- Not yetCash-out to local currency, pay-in-any-token through a Uniswap swap, a security audit, and owner keys behind a multisig and timelock.
- TestnetAll money is testnet money. There is no real loss history, so every return figure here is arithmetic, not evidence.
Demand for credit is not the bottleneck. Acceptance at the till is.
The sales-share model is proven and the finance gap is enormous. What is small today — and growing fast — is merchants taking stablecoins at the counter. The advance is the reason to switch. Nobody changes how they get paid for a slightly better rail; they change it for working capital they cannot otherwise get.
One corridor, 25 real merchants
- Pick one city with existing stablecoin habits and a licensed off-ramp partner.
- Ship the real zkTLS provider for the dominant local seller dashboard.
- Hand-onboard 25 shops with a small pilot pool. Target: a first full repayment cycle with real money.
- Measure what nobody has for this model: actual loss rate, repayment pace, dropout at each step.
Supplier-led distribution
- Go through the people merchants already owe money to: wholesalers and distributors who want paid-on-delivery orders.
- A supplier who gets paid upfront sells the advance for us, shop by shop.
- Standing Pass makes the second advance bigger and faster — retention is the growth engine, not acquisition.
- Open the senior vault to outside lenders once there is a published loss series.
Second market, same playbook
- Repeat in a second country: new zkTLS provider, new off-ramp, same contracts.
- Audit, multisig and timelock before any meaningful outside capital.
- Pay-in-any-token at checkout, so the customer's choice of token stops being the merchant's problem.
What the market looks like if the wedge works
| Layer | Number | Source |
|---|---|---|
| The need | $5.7T small-business finance gap in emerging markets; ~$8T with informal firms | IFC (2019 GDP baselines) |
| The model | Merchant cash advances $20–46B a year, mostly US. Shopify Capital alone advanced $4.2B in 2025 | Mordor, Precedence, Shopify |
| Incumbents in our markets | Moniepoint ~$700M to small businesses in 2025; Safaricom Fuliza now repays from incoming merchant payments | Launch Base Africa, TechTrends KE |
| The rails | LatAm crypto economy $594B, up 9.8% while global flows fell. Cross-border stablecoin transfers up 77.5% in the year to June 2026 | Chainalysis |
| Bottom-up, illustrative | 100k merchants × $200 average advance × ~6 cycles a year ≈ $120M a year advanced | Our assumption, not a forecast |
Three things became true at once.
Stablecoins reached the till
Not as a trade, as money. Cross-border stablecoin transfers are up 77.5% year on year, and the markets where credit is scarcest are the ones adopting fastest.
Proofs got cheap enough for a phone
Noir proofs generate in the browser on a mid-range Android, and zkTLS turns an existing seller dashboard into a credential without an API partnership.
Mandates replaced approvals
World Agent Pass lets a merchant delegate once, with onchain limits, so an agent can pay suppliers without a human tapping approve on every draw.
We also checked the space before building: across 2,824 projects from 2026 ETHGlobal events, no project description matched merchant advances. Credit-scoring projects are everywhere and none reached the finals we could find. We never pitch this as a credit score — it is an advance against a flow we can see.
$400K pre-seed — 18 months to one real repayment cycle.
We are not raising to build the product; it is built, deployed and open. We are raising to find out whether real merchants in one city repay — which is the only number that matters next, and the one number no amount of engineering produces.
| Use of funds · 18 months | USD |
|---|---|
| Team: 3 founders + 1 engineer | 190,000 |
| External security audit | 60,000 |
| Pilot market ops and onboarding | 60,000 |
| zkTLS provider, off-ramp, infra | 40,000 |
| Entity, legal, compliance ×2 markets | 30,000 |
| Buffer | 20,000 |
| Total | 400,000 |
Equity · $400K on a SAFE
Around a $5M post-money cap. A $90K–$500K first cheque fits as lead or co-lead.
Non-dilutive · the privacy roadmap
A ZK non-membership proof for the ban list, an attestor with no trusted issuer, and private writes. Public goods that stand on their own, grant-shaped rather than equity-shaped.
Not equity · $25K pilot lending pool
First-loss capital so 25 merchants can draw real $50–$300 advances. Raised from lenders or a grant — never out of the equity round.
What we will be able to show in 6 months: a published loss series on real advances, time-to-first-advance per merchant, repayment pace, and the dropout rate at every step of onboarding. If the loss rate kills the model, we will say so with the numbers.
Three student builders who shipped the whole stack.
Contracts, Noir circuits, the agents service, the SDK and the app were built and tested by the three of us — 875 offline tests, live on Sepolia, MIT from day one.