Guardrails
You set the systems an agent may touch and the actions it may propose. Outside that brief, it stops.
National Automaton Production AI agents
Production AI agents
AI agents built to take a defined job inside a company — with guardrails, a person to approve consequential actions, and an audit trail of the work.
Each agent is scoped to a task, kept inside that scope, and tied into the systems a company already uses.
You set the systems an agent may touch and the actions it may propose. Outside that brief, it stops.
The agent prepares a consequential step and waits. A person reviews the proposal, and the change goes forward only with that approval.
Every run leaves a record a reviewer can read: what the agent looked at, what it concluded, and what it asked to do.
Integrates with
Ticketing, chat, and mail — including Jira and ServiceNow — so the agent meets the request where it already is.
First product
An AI FinOps agent. It detects cloud cost anomalies, explains the root cause in plain English, and opens a ticket with a proposed fix. A person approves the fix before anything changes.
Open core. The agent framework and MCP tools are open source. Hosted operation, enterprise connectors, and support are paid.
Open source
The agent framework and the MCP tools are the open-source half: readable, runnable, and open to extension.
Paid
Hosted operation, connectors for enterprise systems, and support for teams that want the agents run with them.
The framework and tools can be read and run without the hosted service.
National Automaton builds production AI agents for the enterprise. The name nods to the National Automaton Co., a coin-operated automaton maker in Ohio in 1893. That is inspiration — a machine built to do one job and keep doing it — and this company claims no lineage from that firm.
For the agents, the open-source work, or Spend Automaton.