MAQPNA

Learn

What are guardian agents? The governance layer for AI agents.

As agents take real actions, something independent has to watch them, check them and stop them.

Updated 1 November 2026.

FAQ

Common questions

What is a guardian agent in simple terms?

A guardian agent is something that supervises an AI agent. It watches what the agent does, checks it against rules, and can block or adjust actions that break them. It can be a model, a rules engine or both.

Is MAQPNA named in Gartner's Market Guide for Guardian Agents?

We make no such claim. MAQPNA's gateway works as an in-line, policy-enforcing layer of the kind the guide describes, but you should read the guide itself for Gartner's vendor coverage.

What is the difference between guardian agents and guardrails?

Guardrails usually run inside the application, often in the same process as the model, so a compromised agent may skip them. Guardian agents are meant to be independent of the agent they supervise. Enforcement at a gateway outside the agent's process is one way to get that independence.

Do I need a guardian layer if I already use an AI governance platform?

Usually yes. Governance platforms track models, risks and approvals on paper. A guardian layer acts at runtime, on each call. They complement each other.

A guardian layer you run yourself.

Put every agent action through one deterministic gateway on your own infrastructure, and keep the evidence.

More install options
curl -fsSL https://maqpna.com/install.sh | sh