Search papers, labs, and topics across Lattice.
ChainClaw is a blockchain-native agent framework designed to address critical limitations of large language model agents in on-chain environments, specifically targeting gaps in reactivity, irreversibility, and observability. By implementing a layered architecture that includes an event-driven orchestration layer, a simulation-based safety intelligence layer, and an on-chain monitoring runtime layer, ChainClaw enhances the reliability of tool-augmented tasks in blockchain settings. Evaluation against a benchmark of seven tasks demonstrates that ChainClaw significantly outperforms existing baselines in both safety and task completion metrics.
ChainClaw closes critical gaps in on-chain execution, achieving superior safety and task completion compared to existing frameworks.
General-purpose large language model agents have achieved strong performance on tool-augmented tasks, yet they rely on assumptions break down in blockchain environments. On-chain execution is stateful, adversarial, and economically irreversible, exposing three fundamental gaps: Reactivity, Irreversibility, and Observability. We propose ChainClaw, a blockchain-native agent framework built on OpenClaw, that addresses all three gaps through a layered architecture comprising an event-driven orchestration layer, a simulation-based safety intelligence layer, and an on-chain monitoring runtime layer, unified by a cross-layer memory subsystem. ChainClaw closes the Reactivity gap via event ingestion and simulation feedback, the Irreversibility gap via a pre-execution safety pipeline with transaction simulation and action guard, and the Observability gap via an on-chain read adapter and transaction monitor. We evaluate ChainClaw on a purpose-built benchmark covering seven tasks across four categories and five dimensions. ChainClaw consistently outperforms representative baselines on both safety and task completion.
