Latch: Policy-Bounded Agent Commerce
An agent-native access layer for scarce commerce using paid queues, spend guardrails, and signed reservation state.
Overview & Research Motivation
Scarce tickets, product drops, reservations, and coupons are commonly allocated through manual refresh races and CAPTCHA-style bot blocking. Latch instead models safe agent participation through seller-published paid queues and an autonomous buyer agent operating under explicit spending constraints.
The buyer evaluates live market and Alpha Arcade signals before a deterministic guardrail permits an x402 payment on Algorand. Successful payment advances signed queue and temporary-hold state, ending in a signed claim token or receipt that the seller can verify through the dashboard.
The Core Systems Problem
- Scarce-commerce systems are designed around human refresh races, while autonomous agents need an explicit participation protocol rather than unrestricted checkout access.
- Buyer agents must evaluate signals and respect spend constraints without allowing probabilistic model output to authorize payment or mutate queue state directly.
My Specific Technical Contributions
- Implemented the seller campaign, paid-queue, temporary-hold, and signed claim-token flow.
- Connected the buyer agent to live signal inputs and deterministic spend constraints before x402 payment authorization.
- Integrated x402 settlement on Algorand and signed queue/hold receipts exposed through the seller dashboard.
System Architecture & Verification Pipeline
Seller Campaign & Paid Queue
Sellers publish scarce-access campaigns whose queue and temporary-hold state advances only through verifiable paid participation.
Signal-Aware Buyer Agent
The agent evaluates market signals while deterministic buyer-side spend constraints gate payment authorization.
Algorand x402 & Signed State
x402 payment produces signed queue and hold tokens, followed by a verifiable claim token or receipt for seller fulfillment.
Technical Stack & Tools
Connected Systems & Inquiries
A multi-agent decision architecture where specialized agents deliberate before constrained, user-authorized execution.
An experiment in enforcing cryptographic trust boundaries so generative AI cannot silently redefine canonical game state.
A policy and finance-operations guardrail for x402-enabled autonomous machine payments.