On-chain limits required for AI agent stablecoin payments

Developers built a project called Namera to enforce on-chain spending rules for AI agents that pay in stablecoins. The system uses programmable smart accounts, scoped session keys and on-chain policies to restrict token types, per-payment amounts, daily budgets, approved destinations and session expiry.

Stablecoins now account for a large share of on-chain transaction activity. Adjusted stablecoin volume reached $1.79 trillion in June 2026, and one stablecoin represented about 70% of adjusted volume in the first half of 2026. A web-based payment flow using the HTTP 402 status has processed roughly 165 million agent transactions across about 69,000 active agents by April 2026. Those figures reflect growing automated on-chain payments by software agents.

Most agent deployments today place a private key in an environment variable or config file. In that setup, any spending limits exist only in the agent’s instructions and are not enforced by the wallet. Stablecoin transfers settle quickly, are final and can occur at any hour, so an agent error, a prompt injection, repeated retry attempts or token confusion can move funds before a human can intervene. Keys exposed in logs or shared environments can also give wide access to anyone with access to those systems.

Namera replaces the single-key model with an on-chain permission layer. A user creates a self-custodial smart account controlled with a passkey and issues a scoped session key to an agent. The smart account evaluates policies tied to that session key when a transaction is submitted. Policies can limit which token the agent may use, cap the value of each payment, enforce a daily spending limit, restrict destination addresses or contracts, and set an expiration time. Transactions that violate a policy are rejected by the account, and access can be revoked without moving the underlying funds.

The project includes a dashboard, an API, a TypeScript software development kit, a command-line interface and a local server that supports the agent payment standard. Namera currently supports networks where much agent payment activity occurs, including Base and Base Sepolia, and is being developed as open-source multi-chain infrastructure. Session keys remain on the user’s device and permissions are enforced on-chain rather than relying on agent logic.

Practical policy examples used by developers include a purchasing agent restricted to USDC with a $10 per-call cap and a $100 daily budget limited to approved service contracts; a treasury agent allowed larger per-vendor limits but restricted to vetted vendor addresses; and a trading agent permitted multiple tokens but prohibited from transferring assets to arbitrary external addresses. Standards that enable agent payments at the protocol level exist; Namera focuses on the authorization layer that enforces what an agent may actually spend on-chain.