How Does Curve Llamalend Work?

Permissionless isolated lending markets for borrowing crvUSD against various collateral types

Each lending market is isolated with independent risk parameters. Markets can be created permissionlessly, allowing any collateral type. Isolation prevents cross-market contagion but permissionless creation introduces quality variance.

Lending-Liquidating AMM Algorithm (LLAMMA) providing continuous soft-liquidation for borrowers

LLAMMA continuously converts collateral to crvUSD as price falls and back as price rises, creating soft-liquidation bands. This novel mechanism avoids the cliff-edge liquidation of traditional lending but introduces complexity and potential value leakage through repeated conversions.

crvUSD minted against volatile collateral in lending markets

Borrowers mint crvUSD against collateral like ETH, wBTC, stETH, and other assets. Each market has independent debt ceilings and risk parameters set by governance.

Exponential Moving Average oracle from Curve pool trading data for LLAMMA price feeds

LLAMMA uses Curve's internal EMA oracle for price feeds. The EMA smooths price data but can lag during flash crashes, potentially delaying soft-liquidation initiation.

Lending vaults where users supply crvUSD to earn interest from borrowers

Lenders deposit crvUSD into specific markets and earn variable interest rates determined by utilization. Each market has its own interest rate curve.

Permissionless lending market factory for deploying new collateral markets

Anyone can create new lending markets via the factory. This enables rapid collateral expansion but means poorly researched collateral types can be listed.

Smart contracts written in Vyper, inheriting language-level risk from Curve ecosystem

All Llamalend contracts are written in Vyper, sharing the language-level dependency risk with broader Curve Finance. The July 2023 Vyper compiler exploit demonstrated this risk class.

During sustained multi-week price drops, LLAMMA continuously converts collateral to crvUSD. Borrowers may find their collateral fully converted with extra slippage from repeated conversions, effectively creating a hard liquidation with worse outcome than a single liquidation event.

Permissionless market creation with user-configured oracles enables oracle manipulation attacks. On March 2, 2026, an attacker used a 0M flash loan to inflate the sDOLA exchange rate via a donation attack, forcing 27 borrowers into liquidation and extracting 40K. The vulnerability was in the oracle configuration, not the smart contract code itself.

Internal EMA oracle can diverge from spot prices during rapid moves, but the March 2026 sDOLA exploit showed that external collateral oracles in permissionless markets are the more acute vulnerability — donation attacks can manipulate collateral valuation to force mass liquidations.

Governance-set debt ceilings across multiple markets mean total crvUSD supply can expand rapidly. If multiple markets experience simultaneous bad debt, aggregate losses could pressure the crvUSD peg.

A new Vyper compiler vulnerability would affect all Llamalend markets simultaneously. The July 2023 precedent demonstrated this systemic language-level risk.