Community Guardrails: @sentinelseed/voltagent - AI Safety Integration

[sentinel-seed]

Summary

We've published @sentinelseed/voltagent - a comprehensive AI safety guardrails integration for VoltAgent.

Features

THSP Protocol (Truth-Harm-Scope-Purpose)

• 47 patterns for jailbreak detection, instruction manipulation, and filter bypass
• Four-gate validation system for comprehensive safety checks

OWASP Protection

• 50+ patterns covering SQL injection, XSS, command injection, SSRF, prompt injection
• Real-time threat detection for LLM inputs

PII Detection & Redaction

• 15+ patterns for email, phone, SSN, credit cards, API keys, JWT tokens
• Streaming support for real-time PII redaction

Installation

npm install @sentinelseed/voltagent

Usage

import { createSentinelGuardrails } from "@sentinelseed/voltagent";

const { inputGuardrail, outputGuardrail, streamHandler } = createSentinelGuardrails({
  preset: "standard"
});

const agent = new Agent({
  name: "my-agent",
  model: openai(),
  inputGuardrails: [inputGuardrail],
  outputGuardrails: [outputGuardrail],
});

Links

• npm: https://www.npmjs.com/package/@sentinelseed/voltagent
• GitHub: https://github.com/sentinel-seed/sentinel/tree/main/packages/voltagent
• Docs: https://sentinelseed.dev/docs/voltagent

Metrics

• 186 tests passing
• Full TypeScript support
• Compatible with @voltagent/core >=0.1.0

Would love to have this listed in the VoltAgent ecosystem/community section!