loading
loading
Code that evaluates rules about what an agent is allowed to do before tools or outputs are accepted.
A policy engine turns safety and business rules into enforceable checks: allowed domains, spending limits, PII handling, tool permissions, approval requirements, or content restrictions. Unlike a prompt, it runs outside the model and can block an action deterministically. Example: deny any tool call that would email a customer unless the ticket is linked and approved.
Plainly
Think of Policy Engine as the lock and rules that keep people and data safe. Code that evaluates rules about what an agent is allowed to do before tools or outputs are accepted.
In practice
Use it when a workflow touches secrets, permissions, user data, payments, or untrusted input. In practice, define the owner, input, output, and failure mode before you rely on it.