Skip to main content

FabricZero Documentation

Welcome to FabricZero, the constitutional AI governance platform for autonomous enterprise agents.

What is FabricZero?​

AI agents are software that can take actions on your behalf—sending emails, querying databases, making purchases, writing code. Unlike traditional software that does exactly what it's programmed to do, AI agents make decisions. They interpret goals, choose approaches, and act autonomously.

This is powerful. It's also risky.

An AI agent with access to your company systems is like a new employee with admin credentials on their first day. They might be brilliant, but they don't know your policies, your culture, or your boundaries. Without guidance, even well-intentioned actions can cause harm: sending confidential data to the wrong person, deleting records that shouldn't be deleted, or making commitments the company can't keep.

FabricZero solves this by giving AI agents a constitution.

The Problem: Blind Agents​

Most AI governance today works like an invisible electric fence. The agent tries to do something, gets blocked, tries again, gets blocked again—often entering a frustrating loop. The agent doesn't know the rules exist. It's being punished, not taught.

This creates three problems:

  1. Wasted effort — The agent burns through time and resources attempting things it was never going to be allowed to do
  2. Poor experience — Users watch their agent fail repeatedly with no explanation
  3. No learning — The same mistakes happen over and over because nothing changes

The Solution: Teach, Then Enforce​

FabricZero takes a different approach. Instead of just blocking bad actions, it teaches agents the rules before they start working.

When an agent starts a task, FabricZero injects your organization's values, policies, and boundaries directly into the agent's understanding. The agent knows from the beginning: "I can read customer records, but I cannot modify billing information. I can send emails to one person, but bulk emails require approval."

The agent self-governs because it understands the rules. The enforcement layer exists as a safety net—catching edge cases—not as the primary control mechanism.

Without teaching: The system is a punisher. With teaching: The system is a teacher.

How It Works​

FabricZero operates through three layers:

Layer 1: Knowledge — When an agent starts, it receives your organization's constitution: your values, your policies, what's allowed, what's forbidden, and what requires approval. The agent knows the rules before it takes any action.

Layer 2: Enforcement — Every action the agent attempts passes through a gateway before it executes. Low-risk actions (reading a file) pass through instantly. Higher-risk actions (sending money, deleting data) trigger a quick review.

Layer 3: Debate — For genuinely difficult decisions, FabricZero runs a structured debate. One AI argues for the action, another argues against, and a third makes the final call based on your organization's values. This takes seconds, not hours, and creates a clear record of why the decision was made.

The Result​

Your AI agents become a managed workforce:

  • They know the boundaries — Agents understand what they can and cannot do before they try
  • Every action is reviewed — Nothing executes without passing through governance
  • Important decisions get scrutiny — High-stakes actions trigger real deliberation
  • Everything is auditable — Full records of every decision and why it was made
  • Trust is earned — Agents that consistently make good decisions gradually earn more autonomy

Think of it as the difference between giving a new employee a rulebook on day one versus waiting for them to make a mistake and then firing them. FabricZero gives your AI agents the rulebook.

Key Features​

  • Constitutional Injection — Agents receive your values and policies at startup
  • Scope Enforcement — Agents can only use tools they're delegated
  • Risk Classification — Actions are tiered by risk level for appropriate review
  • Dialectic Review — Important decisions go through structured AI debate
  • Audit Logging — Every decision is logged with full reasoning
  • Trust Ladder — Agents earn autonomy through consistent good behavior
  • Learning Loop — The system gets smarter from every mistake

Getting Started​

Integration Guides​

Reference​

Self-Hosting​

Installation​

Choose your integration:

# TypeScript/JavaScript
npm install @fabriczero/sdk

# Python
pip install fabriczero

# Agent Hooks
npm install fabriczero-agent-hooks

Example​

import { FabricZeroClient, isApproved } from '@fabriczero/sdk';

const client = new FabricZeroClient({
  baseUrl: 'https://api.fabriczero.ai',
  apiKey: process.env.FABRICZERO_API_KEY,
});

const decision = await client.evaluate({
  agentId: 'finance-agent',
  action: {
    tool: 'payment.transfer',
    parameters: { amount: 5000, recipient: 'vendor-123' },
  },
});

if (isApproved(decision.decision)) {
  // Execute the action with the governance token
  await executePayment();
  await client.confirmExecution({
    requestId: decision.requestId,
    token: decision.token,
    success: true,
  });
}

Support​