AgentDock
Policy controlled AI agents that buy risk data without exposing keys or private policy.
Demo
About
AgentDock is a policy-controlled workflow platform for organizations that want AI agents to purchase specialist services and produce recommendations without receiving unrestricted credentials, treasury access, or private policy data. Its first workflow is a private treasury-risk assessment: a delegated agent discovers and pays for an independent risk report, evaluates it against an organization’s confidential risk policy, and produces a minimal verifiable receipt. The project combines scoped, expiring agent identity, metered service payments, confidential policy evaluation, and auditable workflow events. The goal is to make AI agents useful in high-trust financial workflows while preserving human control, revocation, spending bounds, and privacy.
How it's made
AgentDock is built as a TypeScript pnpm/Turborepo monorepo with a Next.js/React dashboard, a Fastify workflow orchestrator, and a separate Fastify x402-gated Risk API. ENSv2 on Sepolia provides organization, agent, and service identities with scoped capabilities, expiry, and revocation. The agent uses Hedera x402 to obtain a paid risk report without receiving unrestricted treasury authority. A Chainlink CRE Confidential Workflow evaluates the paid report against private organizational thresholds inside handlerInTee, returning only a decision and decision hash. The orchestrator persists immutable, idempotent workflow events and writes a minimal Sepolia receipt containing hashes, ENS references, payment reference, CRE decision, status, and timestamp. The dashboard supports onboarding, workflow creation, live event inspection, evidence links, approval/rejection, and agent revocation. Private keys, policy inputs, and service credentials remain outside the agent context and source repository