Money laundering within blockchain ecosystems introduces novel complexities due to the pseudonymous, high-frequency, and cross-chain nature of digital asset transactions. Traditional Anti-Money Laundering (AML) systems face significant limitations in scalability, transparency, and cross-jurisdictional coordination, particularly in contexts involving decentralised finance (DeFi), non-fungible tokens (NFTs), and privacy-enhancing tools. While emerging solutions based on agent-based intelligence, Graph Neural Networks (GNNs), and federated learning offer promise, they remain fragmented and difficult to operationalise. This paper proposes a swarm-based agentic AI architecture that unifies these elements into a scalable, auditable, and regulator-aligned AML system. The framework employs five specialised agents—Placement Detection, GNN-based Flow-Analyser, Graph Modeller, Tracker, and Integration Agents—that collaboratively monitor, record, and analyze transactions across the full laundering lifecycle. All agents commit their findings to a blockchain-based ledger, enabling verifiable decision-making. The Integration Agent specifically monitors whether value cycles back to its origin, flagging such flows as warnings. If this pattern is repeated, the Detector Agent escalates it as a suspicious event. The architecture supports privacy-preserving collaboration via federated learning and delivers explainable insights through graph-native attribution. This integrated approach advances AML capabilities for digital assets, aligning technical innovations with regulatory and operational requirements in a global, decentralised financial landscape.