Nava is a verification layer for AI agent financial transactions. It sits between agent decisions and on-chain execution, using a policy engine to run structured checks on each proposed transaction (transfers, swaps, payments) and return an allow/deny verdict with reasons. Developers integrate via SDK or API, submit agent actions to Nava for verification, then execute or block based on the verdict. Its core mechanism is a graph-of-thought framework that decomposes transactions into check nodes, producing auditable reasoning trails posted onchain, ensuring agent actions comply with preset policies and preventing fund loss from misread instructions or parameter errors. Nava is framework- and model-agnostic, with traceable compliance records.