The three.ws agent engine: a plan/execute decision loop with human-approval gates, a seven-layer GuardChain (security blacklist, intervention, capability, permission, trade guard, spend envelope, x402 budget), a tamper-evident hash-chained action ledger, and a multi-step transaction pipeline with rollback.
nodejs javascript agent open-source threejs sdk human-in-the-loop ai-agents solana guardrails tool-calling llms-txt x402 agents-md agent-runtime decision-loop spend-limit three-ws action-ledger
-
Updated
Oct 10, 2026 - JavaScript