AgentPMT Vision
AgentPMT is the operating layer for autonomous AI agents. Persistent identity, USDC budgets enforced on Base, a credential vault that proxies API calls, and a complete audit trail come built in.
What this looks like
The agent has a 500 USDC budget enforced on Base and calls a Stripe Connected API through the AgentPMT proxy, so the raw key never leaves the credential vault.
By morning three contracts are renewed, two are flagged for review, and every action is searchable by transaction hash, workflow run, or correlation ID.
The gap
The demo is easy to love. The next morning is where trust breaks: who owns the agent, what tools can it touch, what did it spend, which credential did it use, what failed, and can the work run again?
AgentPMT exists for that next morning. It turns one-shot intelligence into work that is scoped, repeatable, inspectable, and accountable on chain.
What changes
How it works
AgentPMT does not replace the model. It wraps every agent run with the parts production systems need: persistent identity, tool access, USDC budgets on Base, a credential vault, approval paths, and an audit trail that survives the session.
Built for trust after the demo
Who this is for
Ship agents that hold an identity, pay for tools, and survive a refresh without rebuilding the plumbing every time.
Hand vendor follow-up, scheduling, and reconciliation to agents you can pause or revoke instantly when something looks wrong.
Set USDC budgets, watch every transaction land on Base, and stop an agent in one click without redeploying anything.
List a tool once, get paid in USDC every time an agent calls it, and skip invoicing entirely.
What this enables
The point is not automation for its own sake. It is fewer draining loops, fewer dead ends, and more room for judgment, craft, family, rest, and invention.
Trust architecture
It is the shape of the system: persistent identity, an encrypted credential vault, USDC budgets enforced on Base, approval paths, append-only audit logs, reusable tools, and human override.
Yes. Pause its smart contract, revoke its API keys, or end its session from the dashboard. Every action takes effect within seconds and is recorded in the audit log.
Agents spend USDC from a smart contract controlled budget on Base. Payments are routed through Circle programmable wallets, capped by the budget you set, and visible on chain by transaction hash.
Every tool call, credential use, payment, and outcome is written to an append-only audit log with correlation IDs. You can search by user, workflow run, agent identity, or transaction hash.
Both. Solo developers ship agent products on top of the infrastructure. Ops, finance, and operations teams delegate operational loops to agents with controls in place. Vendors monetize APIs by listing them as USDC priced tools other agents can call.
An identity, a budget, a set of allowed tools, and the AgentPMT MCP endpoint. Agents call the Model Context Protocol server and the platform handles authentication, spend tracking, credential vault access, and logging.
AgentPMT is the operating layer for AI agents. It gives agents a persistent identity, a credential vault that proxies API calls so agents never see raw keys, USDC payment budgets enforced on the Base network through Circle programmable wallets, and an append-only audit trail every action lands on.
