Averment

Introduction

Averment is a governance and authorization layer for AI actions.

It sits between intent and execution, checking every AI decision in real time using multi-layer intelligence. Only decisions that are safe, authorized, and aligned are allowed to execute.

Beyond safety, Averment gives you full control over how your system behaves.

Works With Your Existing AI Stack

Averment does not replace the AI models you already use.

It has its own internal meaning layer that understands what actions are trying to do. By default that layer uses Averment's built-in AI system. You can point it at your own model instead, and that is the recommended approach.

Your model powers the understanding. Averment handles the governance. These are two separate jobs, and Averment is built to do the second one regardless of which model does the first.

Why Averment

Averment helps you:

Automate more work without losing control
Prevent costly AI mistakes before execution
Reduce prompt injection and unauthorized actions
Govern AI agents using policies and approvals
Protect sensitive systems, data, and resources
Maintain visibility into AI-driven decisions
Apply consistent controls across all AI systems
Use any AI model while keeping the same controls
Customize how AI actions are evaluated and enforced

How Averment Works

Averment sits between intent and execution. Every action your system proposes passes through Averment before it runs.

System / Agent
      ↓
   Intent
      ↓
  Averment
      ↓
 Enforcement
      ↓
  Execution

Averment enables three governance models:

Agent Feedback Loops
Hybrid Human + AI Control
Multi-System Governance

1. Agent Feedback Loops

In this use case, Averment sits between an agent's proposed action and execution. Rather than simply blocking actions, it provides structured feedback that agents can use to modify their behavior and continue operating safely.

Agent
  ↓
Proposed Action
  ↓
Averment
  ↓
Review
  ↓
Agent Replans
  ↓
New Action

This is useful when agents are running autonomously. Instead of failing silently or crashing on a blocked action, the agent reads the feedback from Averment, understands why the action was flagged, and tries a safer approach. The loop continues without human involvement unless the risk is too high.

2. Hybrid Human + AI Control

Not every decision should be fully automated. Some need a human in the loop. Hybrid control lets you define exactly where automation ends and human oversight begins.

An example setup

You define what each outcome does in your system, and Averment enforces it consistently.

3. Multi-System Governance

When multiple agents, services, or workflows all need consistent enforcement, Averment becomes the shared governance layer across your entire system.

An example setup

Every system routes through the same evaluation logic. You define policies once, and they apply everywhere. This removes the need to duplicate enforcement logic across services and makes it easy to audit, update, or tighten controls across your full stack from a single place.

Next steps

Now that you understand what Averment is, get up and running in minutes.

Quickstart

Install the SDK, set your API key, and make your first Averment API call in minutes.