For AI Agents
AI agents are increasingly autonomous, but they can’t prove they’re acting for a real human and shouldn’t be trusted with unilateral control of keys or funds. human.tech treats an agent as a delegated identity: it can act, but every consequential action is gated by its owner’s approval and bounded by policy.
The challenge
Give an agent a private key and you hand it unchecked authority. Give it none and it can’t do anything useful. Meanwhile, services can’t tell whether the agent on the other end represents a verified human or a swarm of bots. The result is the “lethal trifecta”: autonomous capability, sensitive access, and no human gate.
How human.tech helps
- Delegated custody. An agent transacts through a WaaP wallet whose key is split between parties (Two-Party Computation). The agent never holds the full key, and can’t act alone.
- Credential isolation. An agent should not hold raw API keys either. TAP is a credential proxy: the agent calls services by credential name and never receives the secret value, so a prompt-injected agent has nothing to leak.
- Human-in-the-loop by construction. Consequential actions require a verified-human approval (e.g. a biometric proof). Agents propose; humans approve. The protocol enforces it, not a policy doc.
- Scoped permissions. Pre-approved transaction scopes let agents operate smoothly within limits, with human override always available.
- Proof of personhood for agents. Human Passport lets a service confirm an agent acts on behalf of a real, unique human, not a bot farm.
Money and secrets are separate problems and need separate answers. WaaP bounds what an agent can spend; TAP bounds what it can call. An agent that pays for an API needs both.
In practice
- Agentic payments. An agent pays for services or rebalances within owner-set limits, escalating anything outside them.
- Autonomous workflows. Agents run around the clock; the human is pulled in only when judgment or authority is required.
- Civic & enterprise agents. Delegated agents handle routine tasks against verified credentials, gated at every sensitive step.
Build with it
Scaffold a project in one command:
npx @human.tech/create-agent-walletIt generates a working agent for your runtime of choice (Claude, standalone Node, OpenClaw, or Nous/Hermes), wired to a WaaP wallet, including a valid EIP-8004 agent-registration.json so the agent is discoverable by identity registries.
Example recipes
Each is a self-contained, copy-paste tutorial that gets to a running agent in one sitting.
| Recipe | What the agent does |
|---|---|
| Recurring Payments | Subscriptions, payroll, and scheduled transfers within a spend policy |
| Morpho Yield Optimizer | Moves capital across Morpho vaults to chase yield |
| Uniswap v3 Rebalancer | Rebalances concentrated liquidity positions autonomously |
| Snapshot Governance | Votes on DAO proposals under owner-set rules |
| Polymarket LLM Analyst | Trades prediction markets from model-driven analysis |
| CI/CD Agent | Connects its own wallet and runs real transactions on every PR |
Reference
- WaaP for Agents . Programmatic wallet access via CLI for agents and scripts.
- Starter Agents . Pre-built configurations you can launch in minutes.
- TAP docs . Credential setup, connectors, and approval policies.
Talk to us
Building agents that need to act safely on a human’s behalf? Contact us.