Molt.town

This project implements an agent-driven architecture for cross-chain liquidity management that explicitly separates decision-making, execution timing, fund movement, and on-chain behavior into independent layers. The goal is to eliminate reactive, high-churn liquidity automation and replace it with intentional, explainable capital allocation.

At the core is an OpenClaw agent that continuously observes on-chain state such as balances, pool fees, volatility, and estimated execution costs. The agent is initialized with clear objectives and constraints—for example, to maximize net fee yield, avoid frequent rebalancing, and justify every action economically. Rather than executing transactions directly, the agent emits high-level intents (e.g., “move liquidity if expected fee improvement exceeds execution cost”), and frequently decides to remain idle when conditions are not compelling.

These intents are passed to Yellow Network, which buffers, delays, and validates them over time. Yellow ensures that execution only occurs if conditions remain stable, reduces unnecessary on-chain churn, and enables batching or netting of intents. Importantly, Yellow does not move funds or deploy liquidity—it optimizes when execution should happen, not what happens.

When an intent is released, LI.FI handles cross-chain movement by selecting the safest and most cost-efficient bridge and swap routes, optimizing for gas and slippage. Once funds arrive on the destination chain, liquidity is deployed into Uniswap v4, where programmable hooks define how capital behaves—such as dynamic fees, treasury routing, or exit constraints.

The demonstrated strategy moves capital from Base to Arbitrum, where it is deployed into Uniswap v4-native pools to earn fees intelligently. The system produces sparse, well-justified actions with clear logs explaining why liquidity moved, resulting in a transparent, efficient, and truly agentic cross-chain liquidity workflow.

This project is built as a modular, agent-centric system where responsibilities are cleanly separated, but execution authority remains with the agent itself.

At the core is an OpenClaw agent that does hold keys and can execute transactions directly. The agent is configured with a goal-oriented prompt and structured observation inputs such as balances, pool fees, volatility, and estimated execution costs. It reasons about whether an action is economically justified and, only when conditions are met, constructs and signs transactions end-to-end. This makes the agent fully autonomous while still operating under explicit constraints designed to minimize churn and unnecessary execution.

To support efficient and intentional execution, the agent emits intents to Yellow Network, which buffers, delays, and validates those intents over time. Yellow ensures that execution only proceeds when conditions remain stable, reducing gas waste and preventing short-term noise from triggering unnecessary cross-chain movement. Yellow does not move funds itself; it optimizes when execution should occur.

For cross-chain transfers, the agent integrates LI.FI via its SDK to dynamically select optimal bridge and swap routes, abstracting away chain-specific complexity and optimizing for gas, slippage, and safety. Once funds arrive on the destination chain, liquidity is deployed into Uniswap v4 pools, where hooks define programmable behavior such as dynamic fees or controlled exits.

A particularly notable and hacky addition is one-click OpenClaw deployment. We built tooling that allows a fully configured agent—including its wallet, goals, tools, and execution permissions—to be deployed instantly, making it easy to spin up, test, and iterate on autonomous liquidity agents in minutes rather than hours. This combination of autonomous execution, intent-based optimization, and rapid deployment enables a practical and explainable approach to agentic cross-chain liquidity.