Definitions

AI Agents: Autonomous systems performing specific tasks by analyzing data, making decisions, and executing actions. AI agents can be used in many fields, but in crypto they are generally used to determine when and how to carry out optimal crypto trades. They can be customized to all sorts of tasks, from optimal portfolio management to copy trading.

Coincidence of Wants (CoWs): A phenomenon where two (or more) parties coincidentally hold an item or asset that the other wants, and thus are able to exchange directly without the need for an intermediary exchange. In the case of intents, this means that one user’s intent can coincidentally be the opposite of another user’s intents. For example, one intent to swap asset A for asset B and another to swap B for A form a CoW.

Cross-Domain Maximal Extractable Value (MEV): The maximum value that can be captured from arbitrage transactions executed in a specified order across multiple domains. See: Maximal Extractable Value.

DeFAI: This term is used to describe the intersection of decentralized finance (DeFi) and artificial intelligence (AI). In general, DeFAI describes the incorporation of AI tools like AI agents into DeFi practices and protocols.

DeFAI Interactive Search Engine (DISE): Mantis’s proprietary DISE large language model (LLM) is the framework powering personal AI agents on Mantis. DISE is trained specifically with the Mantis SDK framework to make intelligent swaps and execute complex commands across Ethereum, Solana, and beyond.

Intent: An expression of what a user wants to achieve whenever they interact with a blockchain protocol, for instance “transfer X asset from blockchain A to blockchain B” or “trade X asset for Y asset.” Practically, an intent is an off-chain signed message that encodes which state transitions a user wants to achieve. Unlike transactions, intents are partial. Thus, one can think of intents as parts of transactions that require other direct or indirect parts as complements in order to form a final balanced transaction that satisfies all of a user's constraints.

Large Language Model (LLM): This type of AI model is able to understand and even generate natural human language.

Maximal Extractable Value (MEV): The maximal value extractable between one or more blocks, given any arbitrary re-ordering, insertion or censorship of pending or existing transactions (as defined by Obadia et al., 2021).

Multichain-Agnostic Normalized Trust-Minimized Intent Settlement (Mantis): A vertically integrated, optimized intents settlement framework. Mantis is complete with with expression, execution, and settlement.

Solvers: Entities that compete to determine an optimal solution (in the form of a transaction execution pathway) for a user's intent. See: Intent