Home
AI Audit Evidence Compliance Reporting Regulated Industries Regulatory Coverage Improvement Loop Monitoring & Certifications Pricing Plans Security & Trust
ViriSIM
Real-Time Enforcement

:: RUNTIME GUARDRAILS ::

Stop violations
before they reach users.

ViriSIM sits between your AI model and your users — auditing every response at runtime and enforcing block, flag, or allow decisions based on your policies and regulations.

Start Free Trial How It Works
Pre-Gen
Prompt Injection
Post-Gen
Output Verification
3 Actions
Block/Flag/Allow
Sub-8s
Verdict Response
Agent-Ready
LangChain/CrewAI
How It Works

Two intervention points. Total control.

ViriSIM enforces compliance at generation time AND verifies output before delivery.

1

Pre-Generation Guardrail

Your policy is injected into the system prompt before the model generates. The model sees the rules and complies during generation — not after the fact.

"If user input contains PII, then redact it before processing, and do not reproduce any personal data in the output."
AI Model or Agent Generates Response
2

Post-Generation Verification

Before the response reaches the user, ViriSIM audits it against regulations and your policies. The verdict determines what happens next.

ALLOW

Response compliant — deliver to user

FLAG

Needs review — show with warning

BLOCK

Violation detected — refuse output

Composite Decision

Four keys. Your logic.

ViriSIM gives you the leverage to make final decisions. Mix event, verdict, action, and risk score to build your own enforcement logic.

event

System-level classification. Use this for routing and automation.

"fully_compliant_detected"
"review_needed"
"non_compliant_detected"

verdict

Human-readable status. Use this for dashboards and user-facing messages.

"Fully compliant"
"Review needed"
"Non-compliant"

action

Runtime instruction. What your agent should do with the response right now.

"Allow for process to execute"
"Flag for review by team"
"Block dangerous response"

riskScore

Numeric 0-10. Use this for thresholds and conditional logic.

0–2.99 → Low risk
3–5.99 → Medium risk
6–10 → High risk

Your logic, your rules

Example A
"If riskScore ≥ 6, block. Else, allow."

Simple threshold. Ignore verdict. Just use the number.

Example B
"If verdict is 'Review needed' AND riskScore > 4, flag. Otherwise, allow with warning."

Two-key combination. More nuanced control.

Example C
"If event is 'review_needed' OR (verdict is 'Review needed' AND riskScore ≥ 3.5 and < 4.5 ), notify admin, continue response"

Three-key logic. Maximum control over the enforcement flow.

Example D
"If event is 'review_needed' AND (action contains Allow OR Flag) AND (verdict is 'Review needed' AND riskScore ≥ 5), pause and notify admin."

Full 4-key logic. Maximum control over the enforcement flow.

Real-time alerts. Receive push and in-app notifications for audit verdicts, risk score thresholds, and enforcement actions. Know the moment something needs attention.
Policy Sets

You define the rules. We enforce them.

Policies are plain English. No code. No complex configuration. Just tell ViriSIM what to watch for.

PII Protection

"Do not disclose personal information. Redact all email addresses, phone numbers, and financial account numbers."

Data Privacy

Financial Compliance

"Do not provide advice on tax evasion, money laundering, or illegal offshore structuring. Flag content that violates the PCMLTFA."

Finance

Healthcare Safety

"Do not provide medical diagnoses. Always include a disclaimer recommending consultation with a licensed professional."

Healthcare

Bias Prevention

"Do not make assumptions based on race, gender, age, religion, disability, sexual orientation, or national origin."

Fairness

Safety Guardrails

"If the user asks for instructions on illegal activities, refuse and return a compliant response. Do not provide steps or guidance."

Content Safety

Custom Policies

Define any policy in plain English. ViriSIM enforces it at runtime. No limits on the number of policies you can set.

Unlimited
View Sample Policy Set
Agent Tool

One line to add compliance to your agents

Works with LangChain, CrewAI, OpenAI Agents SDK, and custom Python agents.

LangChain

LangChain

Add as a tool in any LangChain agent with one decorator.

@tool
def audit_ai_output(...):
    client = VirisimClient(api_key="...")
    return client.audit(...)
CrewAI

CrewAI

Drop into any CrewAI agent as a compliance guardrail.

@tool("audit_ai_output")
def audit_ai_output(...):
    client = VirisimClient(api_key="...")
    return client.audit(...)
OpenAI

OpenAI Agents SDK

Use as a function tool in OpenAI agent workflows.

@function_tool
def audit_ai_output(...):
    client = VirisimClient(api_key="...")
    return client.audit(...)

Custom Python

Call the ViriSIM API directly from any Python application.

pip install virisim
from virisim import VirisimClient
client = VirisimClient(api_key="...")
result = client.audit(...)
View Python SDK on GitHub
Comparison

How ViriSIM guardrails compare

Most content moderation tools react. ViriSIM prevents.

CapabilityManual ReviewContent Moderation APIsViriSIM
Pre-Generation Control None None Policy injection
Post-Generation Check Slow Basic Full audit
Regulatory Mapping None None 330+ regulations
Custom Policies Manual Keywords only Plain English
Agent Framework Support None None LangChain/CrewAI/OpenAI/Custom
Audit Evidence None None Cryptographically sealed
Response Time Hours/Days Fast Sub-8s
Deploy Method Manual process API only API + SDK + Agent Tool
Sample Guardrail

See a real guardrail from an audit

This guardrail was auto-generated from a detected violation and can be injected into future prompts.

Pre-Generation Guardrail

"If user query implies intent to evade taxes, then refuse and return a compliant response, and do not provide any strategies that could facilitate illegal tax avoidance."

Triggered By

PCMLTFA Part 13 Violation

Action

Refuse + Return Compliant Response

Start building your audit trail today

Get 2,000 free tokens. Run your first cryptographically sealed audit in under 30 minutes.

Start Free Trial

No credit card required. Enterprise plans available.