Find common PII
Detect email addresses, credit-card-shaped numbers, API-key formats, IP addresses, phone numbers, and social security numbers.
Local-first prompt scanning
Catch sensitive data before it reaches your AI.
Prompt Guard checks text for common personally identifiable information and prompt-injection patterns, right in your application or from the command line. Its optional local vault keeps secret values encrypted and out of model-facing tool results by default.
Scan a prompt file
$ prompt-guard examples/clean.txt ✅ Text is clean and safe. $ prompt-guard examples/leak.txt 🚨 Security risk detected!
A practical first line of defense
Run checks before submitting prompts to a model. Scanning happens locally; prompt content is not sent to an external service by Prompt Guard.
Detect secrets
Catch OpenAI keys, GitHub tokens, AWS keys, Stripe keys, database URLs, JWTs, and SSH private keys.
Flag injection phrases
Identify common prompt-injection language, including attempts to ignore previous instructions or expose a system prompt.
Keep credentials local
Store secrets in an encrypted local vault and expose only IDs and safe metadata to harness tools. Detection is best-effort; review captured prompts before sending them to a model.
Pattern-based scanning is best-effort, can produce false positives or miss novel formats, and is not a substitute for application-specific security controls. Injection phrases are reported but not removed from the sanitized text.
Quick start
Install from npm or clone the repository.
Install from npm
npm install -g @ai-clyde/promptguard
Link and scan
npm link prompt-guard examples/clean.txt
Use in your app
import { scanText } from '@ai-clyde/promptguard';
const result = scanText(prompt);
Read the setup guide or usage guide for complete instructions. See the README for development instructions.