In recent years, artificial intelligence (AI) has taken the world - and the blockchain industry- by storm. AI has already served to improve the blockchain space in a number of ways. Now, with Composable’s Mantis cross-chain intent settlement framework, AI agents can deliver even further optimized transaction execution.

Artificial intelligence is a massive sector of technology, and only growing. For example, the size of the AI market is currently around $184 billion, and its annual growth rate (CAGR) is predicted to be 28.46% between 2024 and 2030 (Statista). AI is certainly making its mark in blockchain and DeFi, delivering optimizations in areas like data management, transaction efficiency, and energy consumption. In fact, the market size of blockchain-powered AI is predicted to be over $980 million in 2030 (Spherical Insights). One incredibly popular example of AI in decentralized finance is crypto trading bots (such as those found on Telegram), which automate and optimize transactions. Other examples include AI-based protocols and applications.

We are aiming to make Mantis’s cross-chain intent settlement capabilities leverageable by various types of AI agents. Partner protocols’ existing AI tech stacks can be used to generate these AI agents. AI agents can then be used to submit intents on behalf of their users at optimal times to the Mantis framework.

There are near limitless possibilities for how AI agents can leverage the Mantis framework. Our suggestion is that these agents grab on-chain data in order to make recommendations to users for what intents they should submit through Mantis. The flow of this would look as follows:

1. A user submits their profile/wants to an AI agent.

Example: The user wants to get more involved in the Solana ecosystem. So, they are looking for opportunities to swap their Ethereum-based assets for Solana-based assets at the best price.

1. AI agents monitor on-chain data for opportunities that fit a user’s profile or wants.

2. When the AI agent identifies an opportunity fitting the user’s desires, they present it to the user. In fact, the AI agent can make a number of recommendations at once and present these simultaneously.

Example: The AI agent has found that it is a good time to trade ETH for either SOL or jupSOL, based on the current market conditions.

1. The user accepts the recommendation.

Example: The user opts to swap ETH for jupSOL.

1. The AI agent submits an intent to Mantis on behalf of this user reflecting the selected recommendation. This submission is done using the Mantis Intents SDK (described below).

2. The Mantis network of solvers competes to settle this intent with best execution, sending proposed solution routes to the Mantis auctioneer to be scored based on how well they fit the intent.

3. The winning solution route is determined, and the winning solver must execute this route in order to be rewarded. The Inter-Blockchain Communication (IBC) Protocol is used for cross-chain settlement.

Example: The user’s ETH on Ethereum is sent to Solana over IBC and then swapped for jupSOL.

Using the above process, AI agents become capable of handling time-based conditional intents. For example, a user may want to make a particular swap at the best price within the next 48 hours. They provide this information to the AI agent, which makes a prediction about when the best price will be within this time constraint. Then, at this time, the AI agent submits an intent to carry out the swap to Mantis. . This abstracts conditionality away from solvers and puts it in the hands of AI agents who likely have more powerful algorithms to determine the best timing of swaps. Then, solvers are left to handle identification and execution of the best transaction route at that time. In this manner, the strengths and optimizations of the Mantis protocol and of AI agents are able to synergize, providing the best execution.

Mantis can be accessed by any application via its intents software development kit (SDK). This SDK interacts with the expression layer of Mantis where users express their intents and these intents are received and interpreted by solvers. The intent SDK allows intent expression to be generalized, so Mantis can process and execute intents from many sources. These sources thus leverage the Mantis/Picasso architecture and become intent-centric themselves. In this manner, Mantis acts as back-end processing and chain abstraction for AI agents. This in turn enables support for AI agents to create intelligent transaction routes.

AI agents and protocols are able to tap into Mantis by pointing to the remote procedure call (RPC) of Mantis, as shown in the diagram below:

This will let AI agents submit user intents at strategic times to achieve best execution, with transactions enabled over all IBC-connected chains. This integration can even be directly integrated into the Mantis rollup on Solana.

If an AI protocol integrates Mantis through the intents SDK, it will gain the following value adds:

• Revenue Share: Orderflow from a unique location will receive a  revenue share based on order flow origination. This implies that any AI agent directing unique order flow to Composable automatically receives that revenue share as part of the partnership, once Composable processes the given order.

• Cross-Chain Processing: Mantis serves as a processing back-end for all things cross-chain. Thus, AI agents partnering with Mantis can have all sorts of cross-chain operations processed for them.

Users of these AI protocols integrating with Mantis gain the following benefits:

• Intent-Centricity: These AI agents will be able to process intents, not just transactions. Unlike transactions (which explicitly specify exact steps to take), intents are flexible. They describe a general goal but need other direct or indirect operations in order to form a final balanced transition that satisfies all of the user’s constraints. Intents can be conceptualized as a means by which users express their preferences across various domains without specifying the methods to achieve desired outcomes. Thus, intents offer the possibility of optimized execution for users.

• Conditional Intents/Intelligent Transaction Routes: AI agents will be able to create intelligent transaction routes, which even further improves execution for users. For example, AI agents can submit user intents when prices are optimized (e.g. submitting a buy order when prices are low). This will provide best execution to intents submitted by/through these AI agents, even further optimizing their transactions.

If you have an AI protocol and are interested in integrating Mantis, you can reach out to us on X or Discord.