Agent PolicyPortable policy decisions for agent systems.

Define how agents ask, receive permission, evaluate risk, retain data, redact output, and explain policy outcomes across runtimes and products.

Read the specification

What is Agent Policy?

LLM full context

Policy decisions

Stable facts for allow, deny, ask, defer, escalate, waive, not applicable, and indeterminate outcomes.

Approvals and grants

Human and automated approvals are modeled separately from runtime execution and UI rendering.

Risk and data controls

Risk scopes, redaction, retention, export obligations, credentials, network, filesystem, and remote-agent boundaries.

Audit-ready traces

Every decision can point to policy sets, matched rules, inputs, obligations, waivers, grants, telemetry, and evidence refs.

Start here