Human intent
Let an agent spend within understandable limits while preserving hard rules, soft preferences, uncertainty, exceptions and explicit human approval.
← All projects02 / MONEY / AUTHORITY / 2026
haam-payHuman control for agent payments.
Working prototype with live testnet laboratory
An authority layer between an AI agent and payment execution. It does not move money itself. It decides whether a proposed economic action is within the human mandate.
System / agent action
Evaluate a proposed payment or trade and return ALLOW, ASK or DENY before any signing or execution adapter becomes reachable.
Hard boundary
Payment rails, wallets and trading tools remain downstream. The evaluator sits before signing, maintains independent caps, supports revocation and records sanitized evidence without storing private keys or raw signatures.
INTENT PRESERVATION / CORE
Payment rails, wallets and trading tools remain downstream. The evaluator sits before signing, maintains independent caps, supports revocation and records sanitized evidence without storing private keys or raw signatures.
Evidence in the system
- The same core evaluator gates x402 Exact and Upto test payments before authorization is signed.
- A Kraken integration is intentionally paper-only and exposes no live order command.
- Example mandates include asking before subscriptions and allowing bounded purchases only above a confidence threshold.