Audrey — The Identity and Control Layer for AI Agents

The Identity and Control Layer for AI Agents in Production Systems

Agents are moving from "chatting" to "acting." Production systems respond by throttling, blocking, and flagging machine-shaped behavior. Audrey makes autonomous actions attributable, governable, and auditable.

Join Waitlist Read the Blog

v.01 / Pilot Phase

01 / Overview

WHAT WE DO

No. 012025.01.06

What Audrey Does

Audrey is a governance layer that lets AI agents act on behalf of humans safely. We bind actions to accountable operators, enforce runtime policies at the moment of execution, and generate cryptographic action receipts that downstream systems can trust.

No. 022025.01.05

The Problem

AI agents trigger anti-abuse and fraud defenses because autonomy looks like bot behavior at scale. Attribution breaks across vendor boundaries, rate bursts and retries resemble abuse, and audit trails are fragmented or non-verifiable. Legitimate automation gets labeled, throttled, blocked, or forced into brittle workarounds.

02 / Phases

ROADMAP

From voice deliverability to general agent authorization. Built for production systems.

Phase 1

Voice Firewall

Policy-gated calling with pacing, warm-up ramps, concurrency caps, retry backoff, time windows, anomaly kill switches, and signed per-call receipts.

Phase 2

Delegated Authorization

A general runtime authorization and policy engine for high-risk agent actions. Provable delegation, enforceable constraints, and verifiable audit trails.

03 / Principles

THE PRIMITIVES

Three pillars

Delegation

Scoped, time-bound, revocable authority tied to an accountable operator.

Enforcement

Policy decisions at the point of execution: allow, deny, throttle, or step-up.

Receipts

Signed, reconstructable records: who authorized, what policy applied, what happened, and why.

04 / Waitlist

JOIN THE WAITLIST

Get notified when pilots open. Early access, product updates, and pilot invites.