What awzy is

awzy decides whether one principal may take one action on one resource, every time it happens. It implements the “Beyond Zero” model from Google’s paper of that name: zero trust stops at the front door, so authorization has to follow every action after it, for people and for the AI agents that now act on their behalf.

Early

awzy is in active development and not yet generally available. These pages describe what’s built today; the platform plan says what’s coming.

awzy works alongside the authorization you already have: your IdP, your cloud’s IAM, and your apps’ own permissions. See awzy and your existing authorization for what it adds and how to adopt it a step at a time.

The pieces

Service	What it does
awzy-pdp	The policy decision point. Answers Check(principal, action, resource) with allow, deny, or challenge, in milliseconds.
awzy-gateway	An enforcement point you don’t have to code: a proxy that checks every MCP tool call.
awzy-policy	Versioned, signed Cedar policy bundles: drafts, dry runs, publishing, and break-glass rollback.
awzy-context	Who and what: people and groups from your IdP over SCIM, registered service accounts, and resources with their sensitivity.
awzy-challenge	Challenges (justification, security key, approval) and containments.
awzy-intake and awzy-reasoner	Take in activity from app and audit logs, and score each person’s and agent’s risk from it.
Client portal	Where your security team reviews decisions, publishes policy, answers challenges, and manages the directory.

How it runs

Everything runs in your own cloud account, one install per client: awzy never sees your data, and nothing about your people or resources leaves your install. See the security model and installing awzy.