The rise of the self-custodial neobank: 47 apps where you hold the keys
Every bank in history has worked the same way: you give them your money, they give you promises. The third wave of neobanks breaks that contract. In a self-custodial neobank, the company never holds your funds — your balance is stablecoins in a wallet only you control, and everything bank-like (a card, an IBAN, yield, payments) is software wrapped around it.
We track 47 web3-native neobanks. Of the full 357-entity directory, 40 are fully self-custodial and 2 use MPC self-custody. Two years ago most of these didn't exist. This post is about what they actually are, why now, and where the model still leaks.
From wallet to bank in three steps
The category emerged from two directions converging:
- Wallets grew banking features. MetaMask, Phantom, Trust Wallet, Exodus and Xverse started as key managers for crypto users. Then came in-app buy/sell ramps, then debit cards (MetaMask and Phantom both ship them), then yield. The wallet became the account.
- Banking products grew self-custody. Gnosis Pay built a Visa card on a Safe smart account; Payy and EtherFi Cash followed with their own takes; apps like Peanut and Daimo made stablecoin payments feel like Venmo. The account became a wallet.
The meeting point is a phone app where you can receive a salary in USDC, earn on it, tap a card at a supermarket, and send money to a friend — while the operating company holds nothing and, in the cleanest designs, can't touch your funds even if compromised or coerced.
How the pieces work
| bank feature | self-custodial equivalent | who provides it |
|---|---|---|
| Account | Wallet / smart account (often a Safe) | You + open-source contracts |
| Balance | Stablecoins (USDC, USDT, EURe…) | Regulated issuers |
| Debit card | Card programme with on-chain spending allowance | Licensed issuer + network |
| IBAN / account number | Named virtual IBAN routing to your wallet | Banking-rails partner |
| Interest | On-chain yield (lending markets, T-bill-backed tokens) | DeFi protocols / RWA issuers |
| Fraud protection | Spending limits + module permissions in the smart account | Contract logic |
What you gain
- No custodian to fail. The Synapse collapse froze customer funds for months (that story here); FTX vaporised them. A self-custodial balance cannot be frozen by a bankruptcy trustee, because it was never part of anyone's estate.
- Global by default. A wallet doesn't care about your passport. For people in high-inflation or capital-controlled economies, apps like MiniPay (on cheap Android phones, running on Celo) or El Dorado in Latin America are dollar accounts that no local institution can gate.
- Yield without a middleman's cut. The spread a bank keeps between what your deposit earns and what it pays you is the oldest business model in finance. On-chain, you hold the earning asset directly.
Where the model still leaks
Honesty section — the risks don't vanish, they move:
- Key loss is unforgiving. Seed phrases are terrible UX. The better apps now use passkeys, MPC (Zengo has no seed phrase at all), or social recovery — but "I forgot my password" still has no 1-800 number.
- The edges are custodial anyway. The moment you touch fiat — the card issuer, the IBAN provider, the on/off-ramp — a regulated intermediary appears, with KYC and the power to decline. The core is trustless; the perimeter isn't.
- Stablecoin issuer risk. Your "dollars" are Circle's or Tether's liabilities. The GENIUS Act and MiCA hardened reserve rules considerably, but issuer risk is now the systemic question of this whole category.
- Smart-contract risk. Audited, battle-tested contracts like Safe carry low but nonzero risk. Newer programmes carry more.
- No chargebacks. Push payments on-chain are final. Consumer protection here is genuinely worse than a Visa dispute — card programmes reintroduce some of it, wallet transfers have none.
Why this category will matter
The bear case says self-custody is a niche for the crypto-native 1%. The bull case runs through three doors: emerging markets (where a self-custodial dollar account beats the local alternative on day one — see where neobanks actually win), AI agents (software that needs to hold and spend money can't pass bank KYC, but it can hold keys — this is why we track agent wallets in the directory), and the simple fact that the custodial waves keep having accidents. Every freeze, every bankruptcy, every "temporarily suspended withdrawals" headline is marketing for wave three.
Speculative, flagged as such: we expect the traditional/web3 boundary in our own dataset to blur within two years — traditional neobanks are already adding stablecoin support (103 of 357 have some), and several web3-native apps are acquiring the licences that would technically reclassify them. The categories will converge before the vocabulary does.