AI Governance

Runtime guardrails for
production AI

Intercept AI outputs before they become actions. Evidence validation, confidence scoring, risk assessment, and policy enforcement for regulated industries.

View on GitHub Explore Features
5
Policy Types
Real-time
Enforcement
4
Risk Levels
Evidence
Based Decisions
Capabilities

Everything you need to govern AI

A comprehensive runtime governance layer built for production environments in regulated industries.

Evidence Validation

Verify supporting evidence for every AI decision. Ensure outputs are grounded in factual, traceable data before they reach downstream systems.

Confidence Scoring

Quantify certainty levels for every decision. Threshold-based routing ensures low-confidence outputs are flagged or blocked automatically.

Risk Assessment

Four-tier risk classification from minimal to critical. Each level triggers appropriate review workflows and enforcement actions.

Policy Engine

Configurable policy framework with five built-in types. Define custom rules that match your organization's compliance requirements.

Audit Trail

Complete decision logs with full traceability. Every interception, evaluation, and action is recorded for compliance and forensic review.

Real-time Interception

Sub-millisecond evaluation pipeline. Intercept and assess AI outputs in real-time without adding perceptible latency to your application.

How it works

Architecture

Every AI output passes through the decision firewall before reaching downstream systems.

AI Output
→
Decision Firewall
→
Evidence Check
Confidence Score
Risk Assessment
Policy Match
→
Allow
Block
Escalate
Governance

Policy Types

Five built-in policy types cover the most common governance requirements for regulated AI systems.

Confidence Threshold

Block decisions below minimum confidence levels

Evidence Required

Mandate supporting evidence for all outputs

Risk Ceiling

Set maximum acceptable risk levels per domain

Human Review

Escalate high-risk decisions to human reviewers

Domain Restriction

Limit AI actions to authorized domains only

Quick Start

Get started in minutes

Install, configure, and start enforcing governance policies on your AI outputs.

# Install from PyPI
pip install ai-decision-firewall

# Or install from source
git clone https://github.com/BabyChrist666/ai-decision-firewall.git
cd ai-decision-firewall
pip install -e .
from ai_decision_firewall import DecisionFirewall
from ai_decision_firewall.policies import ConfidencePolicy, EvidencePolicy

# Initialize the firewall
firewall = DecisionFirewall(
    policies=[
        ConfidencePolicy(min_confidence=0.85),
        EvidencePolicy(require_sources=True),
    ]
)

# Evaluate an AI decision
result = await firewall.evaluate(
    decision=ai_output,
    context={"domain": "healthcare"}
)

if result.action == "allow":
    execute(ai_output)
elif result.action == "escalate":
    send_to_reviewer(ai_output, result.reason)
else:
    log_blocked(ai_output, result.reason)
from ai_decision_firewall.policies import BasePolicy
from pydantic import BaseModel

class CompliancePolicy(BasePolicy):
    """Custom policy for regulatory compliance."""

    name: str = "compliance_check"
    restricted_terms: list[str] = []

    async def evaluate(self, decision, context) -> PolicyResult:
        # Check for restricted content
        for term in self.restricted_terms:
            if term in decision.content.lower():
                return PolicyResult(
                    action="block",
                    reason=f"Contains restricted term: {term}",
                    risk_level="high"
                )

        return PolicyResult(action="allow")
Python FastAPI Pydantic asyncio

Govern your AI, responsibly

Start enforcing runtime governance on your AI systems today. Open source and built for production.

Get Started Contact