tracerney

prompt-injection defense
docsgithub

install (free)

$ npm i @sandrobuilds/tracerney

Free SDK with access to 258 real-world attack patterns. Detects suspicious prompts locally with <5ms latency. No backend required. No data collected.

quick start (layer 1)

Get up and running in minutes. The SDK analyzes prompts locally against 258 patterns. Flags suspicious content instantly. Zero network overhead.

const tracer = new Tracerney();
const result = await tracer.scanPrompt(userInput);
if (result.suspicious) {
console.log("⚠️ Suspicious:", result.patternName);
}

No data leaves your server. All detection happens locally with <5ms latency. Simple result object makes it easy to handle suspicious prompts your way.

add layer 2 for defense-in-depth (pro)

Upgrade to Pro to add Layer 2 LLM Sentinel for output validation. Verify that LLM responses haven't been compromised by injection attacks. Industry best practice for high-security applications.

const tracer = new Tracerney({
apiKey: process.env.TRACERNEY_API_KEY,
sentinelEnabled: true,
});
try {
const result = await tracer.scanPrompt(userInput);
return result; // Safe to use
} catch (err) {
if (err instanceof ShieldBlockError) {
return NextResponse.json(
{ error: "Input flagged as suspicious" },
{ status: 400 }
);
}
}

Custom Layer 2? Self-host or use your own server by setting baseUrl and sentinelEndpoint. See docs for details.

quick setup (with coding agent)

Not sure how to integrate Tracerny? Ask your coding agent (like Claude Code) to set it up for you. Your AI assistant can read the npm package documentation and automatically configure the SDK in your project.

How it works:

  1. Install the package: npm install @sandrobuilds/tracerney
  2. Point your coding agent to the npm docs: @sandrobuilds/tracerney
  3. Ask to set up Layer 1 or Layer 2 configuration
  4. Your agent generates the complete implementation

Whether you need basic pattern detection or advanced Layer 2 with custom endpoints, your coding agent can read the docs and set everything up in minutes. No manual configuration needed.

learn

what is injection?

Prompt injection is a technique where an attacker redirects an AI system from its original instructions by injecting malicious input.

read more →

how it works

Tracerny analyzes every prompt against 258 embedded patterns in real-time with threat scoring.

read more →

faq

Common questions about Tracerny: performance, integration, data privacy, and more.

read more →

pricing

free

  • ✓ 258 attack patterns
  • ✓ local detection
  • ✓ 0ms latency
  • ✓ 50 scans/month with layer 2 (testing, early-stage, hobby)
  • ✓ community support

pro

$9/month

  • ✓ everything in free
  • ✓ layer 2 llm sentinel
  • ✓ 2,500 scans/month with layer 2
  • ✓ context-aware scan, llm sentinel output safe json, delimiter salting
  • ✓ live threat dashboard
  • ✓ priority support

frequently asked questions

payments & subscriptions

Do you offer free tier?

Yes! Free SDK with 258 patterns and 50 Layer 2 scans/month. Perfect for testing and early-stage projects.

What payment methods do you accept?

We use Stripe for all payments. Credit cards, debit cards, and other major payment methods are supported.

Can I cancel anytime?

Yes, cancel your subscription anytime from your account. No long-term contracts or hidden fees.

Do you offer refunds?

We offer a 30-day money-back guarantee. If you're not satisfied, we'll refund your subscription.

layer 1 vs layer 2

What's the difference?

Layer 1: Local pattern detection (free). Analyzes user input against 258 embedded patterns. <5ms latency, no network calls.

Layer 2: Server-side LLM Sentinel (pro). Validates LLM outputs with comprehensive schema checks:

{
"action": "ALLOW" | "BLOCK",
"confidence": number, // 0-1 confidence score
"class": string, // safe_content, jailbreak_llm_detected, etc
"fingerprint": string // Unique ID for deduplication
}

Do I need Layer 2?

Layer 1 provides strong protection for most cases. Layer 2 is industry best practice for high-security applications. Recommended when: handling sensitive data, regulatory compliance required, or extra confidence needed.

Why use both layers?

Defense-in-depth strategy. Layer 1 catches 99% of attacks early (fast, local). Layer 2 validates LLM outputs as a second checkpoint. Together they provide comprehensive protection: Layer 1 stops inputs → Layer 2 validates outputs.

general

Is my data private?

Layer 1 is completely local (no data leaves your device). Layer 2 uses end-to-end encryption and we never store your prompts. See our privacy policy.

Is this production-ready?

Yes. Tracerny is used in production by companies handling sensitive data. Layer 1 has <5ms latency and zero external dependencies.

What LLMs does this work with?

Works with any LLM: GPT-4, Claude, Gemini, Llama, or custom models. Language-agnostic pattern detection.

Layer 2 uptime?

99.9% SLA. Runs on enterprise-grade infrastructure with automatic failover.

security & philosophy

Is this 100% unhackable?

No. In security, "100%" is a marketing lie. Tracerny is built on the principle of Exponential Friction. We use a three-layer defense: Slicer, Judge, and Salted Kernel. This makes the cost of a successful injection higher than the value of the data being protected. We don't promise a ceiling; we provide a sentinel.