A standalone MIT-licensed kit (https://github.com/yashishkhurana/arc-boring-vault-agent-kit) providing a model-agnostic merkle-bounded agent guardrail plus an on-chain reasoning journal for AI-managed vaults on Arc. It wraps a full Veda/BoringVault Foundry stack (BoringVault, ManagerWithMerkleVerification, Teller, Accountant, decoders) deployed to Arc with addresses pinned in config, an AgentJournal reasoning log, and an OpenAI-compatible LLM strategist (DeepSeek, with a deterministic rule-engine fallback) that selects from a merkle-allowed action menu re-verified on-chain; a keeper reads the vault's real USDC balance and broadcasts manageVaultWithMerkleVerification plus journal.record.
I'm building @lucidlyfi
2 — Yashish Khurana - founder, Lucidly Labs (x.com/lucidlyFi) Samarpan Dutta - founder, Lucidly Labs [Institutional Execution for automated onchain strategies]
Battle Vaults: two AI agents run competing vaults on Arc. A free LLM picks allocation by market regime, every decision's reasoning is journaled onchain, and a merkle guardrail means the agent can't act outside its pre-approved set. USDC settled, USYC yield, CCTP cross chain. Adaptive Portfolio Manager.