Problem
Liquidations, collateral swaps, refinancing and cross-venue arbitrage all need capital for a moment and none after it. On Solana that moment must fit inside onetransaction: 1,400,000 compute units, 1,232 bytes and 64 account locks for the borrow, the route and the repay together. Existing flash loans live inside lending markets as a side feature, and the repay check — the part that keeps depositors whole — is easy to get subtly wrong.
Design
- Reserves. One vault per asset (SOL, USDC, USDT, JitoSOL, JUP, JTO). Depositors receive shares; the share price only rises, by the fees repaid.
- Borrow. A top-level instruction that walks the Instructions sysvar forward from its own index and requires exactly one matching repay — same reserve, mint and borrower. Stack height above 1 (a nested call) is rejected.
- Repay. Reads the matched borrow’s amount, pulls principal plus fee, and closes the loan. A transaction with a borrow and no valid repay cannot succeed, so it cannot change state.
- One loan per transaction. Removes index-reuse and nested-loan ambiguity entirely.
Economics
The fee is 1 basis point of principal, paid in the borrowed asset, 100% to depositors. Published flash fees run from 0% (marginfi, Jupiter Lend, Morpho, Balancer V3) through 0.001% (Kamino, per reserve) to 0.05% (Aave V3 default). Atomic arbitrage lives on a few basis points of edge, so a fee in tenths of a percent prices most routes out; one basis point does not. Depositor yield therefore comes from turnover, not from a high rate.
$WEDESA
$WEDESA governs the parameters that are deliberately not hard-coded: which assets get a reserve, per-reserve borrow caps, the freeze switch, and the treasury that funds audits and the bug bounty. It is not needed to borrow, and the flash fee is never paid in it — fees stay with depositors in the asset that was lent.
The contract address is published on this site and in the footer at launch. Anything else claiming to be $WEDESA is not.
Risks
- Program risk. A bug in the repay check could let a loan leave without repayment. Mitigations: one-loan-per-transaction, nested-borrow rejection, published kit tests for every revert path, and a public bug bounty.
- Token edge cases. Token-2022 transfer fees and hooks change the amount that arrives; reserves for such mints account for the post-hook amount.
- Network changes. Transaction formats and limits are evolving (larger v1 transactions, higher block compute limits). The SDK tracks them; integrators should pin versions.
- Governance. A captured vote could list a bad reserve. Listing has a timelock and the freeze switch is scoped per reserve.
