Fetch.ai, a Cambridge, U.K.-based artificial-intelligence lab with a penchant for crypto, has launched a service to combat the risk of losses across the experimental decentralized finance (DeFi) space.

Announced Tuesday, the DeFi Agents toolkit can be set to automatically withdraw users’ funds from Uniswap v2 and PancakeSwap based on predefined conditions such as the exchange rate for a given token dropping to a certain level.

While cryptocurrency prices are weakening, DeFi innovation is booming, with lots of insurance-like projects and even DeFi-focused blockchain analytics in the offing.

Fetch.ai, which has applied its machine-learning chops to enterprise blockchain as well as public crypto, is preventing losses associated with decentralized-exchange (DEX) trading and the use of automated market makers (AMMs) by liquidity providers, the participants who deposit tokens on DeFi platforms in order to earn yields.

Currently, the DeFi Agent tool provides “stop-loss agents,” which can automatically withdraw user funds from liquidity pools. These kick in if the exchange rate between the two tokens falls to a predetermined level, according to a press release. In traditional trading, a stop-loss order is a parameter that triggers a purchase or sale of a specific asset once it reaches a given price. At launch, users will be able to create up to five agents with stop-loss triggers for all liquidity pools on Uniswap and PancakeSwap, a Binance Smart Chain-based DEX.

The Fetch.ai team plans to extend the functionality of the DeFi Agent tool to enable automatic liquidity withdrawal and deposit when token sentiment goes below a certain threshold; to move liquidity of ERC-20s or BEP-20s to a defined range if the price is breached (in Uniswap v3); and to remove liquidity if ETH fees are becoming too high in a given period of time, according to the release.

“Intelligent automation has the potential to transform the end-to-end experience of the DeFi applications we use today,” Humayun Sheikh, CEO of Fetch.ai, said in a statement, adding: