Stop shipping unsafe AI output. Local-first verification, safety monitoring, and guardrails for any LLM — in one npm install.
100% Local | No Telemetry | Privacy-First | 500 free calls/day
- Product Page: subodhkc.com/products/llmverify
- npm: npmjs.com/package/llmverify
- GitHub: github.com/subodhkc/llmverify-npm
- Python Package (PyPI): github.com/subodhkc/llmverify
- Author: Subodh KC — AI governance, compliance, and security leader
You're building with GPT-4, Claude, Gemini, or any LLM. Your AI:
- Hallucinates facts and citations that don't exist
- Leaks PII — emails, phone numbers, SSNs in responses
- Gets prompt-injected — users trick it into ignoring instructions
- Returns broken JSON that crashes your parser
- Drifts in quality over time without anyone noticing
You need a safety layer between your LLM and your users. That's llmverify.
npm install llmverify
| Feature | One-liner | What it catches |
|---|---|---|
| Verify output | await verify(aiResponse) |
Hallucinations, harmful content, quality issues |
| Block injections | isInputSafe(userMessage) |
Prompt injection, jailbreaks, malicious input |
| Redact PII | redactPII(text) |
Emails, phones, SSNs, credit cards, API keys |
| Fix broken JSON | repairJSON(broken) |
Missing brackets, trailing commas, unquoted keys |
| Monitor health | monitorLLM(client) |
Latency spikes, token drift, behavioral changes |
| Sentinel tests | sentinel.quick(client, model) |
LLM regressions before users see them |
| Classify output | classify(prompt, response) |
Intent detection, content categorization |
| Audit trail | audit.log(event) |
SOC2, HIPAA, GDPR compliance evidence |
Everything runs locally. No data leaves your machine. No API keys needed for free tier.
Part of the HAIEC AI governance platform. Use alongside AI Security Scanner, CI/CD Pipeline, and Runtime Injection Testing.
npm install llmverifyconst { verify, isInputSafe, redactPII } = require('llmverify');
if (!isInputSafe(userMessage)) {
return { error: 'Invalid input detected' };
}
const aiResponse = await yourLLM.generate(userMessage);
const result = await verify(aiResponse);
if (result.risk.level === 'critical') {
return { error: 'Response failed safety check' };
}
const { redacted } = redactPII(aiResponse);
console.log(redacted);Three lines of safety between your LLM and your users.
- Issues: GitHub Issues
- Docs: Full Documentation
MIT License — See LICENSE for details.
Made with care for AI safety and developer experience.