LLM Guard
ActiveDescription
The security toolkit for LLM interactions, providing prompt injection detection, PII anonymization, content safety auditing, and more to secure production LLM deployments.
Key Features
- Comprehensive prompt scanners: prompt injection, toxicity, secrets, sentiment, and more
- Output scanners: bias detection, factual consistency, malicious URL detection, and sensitive data filtering
- PII anonymization and deanonymization for data leakage prevention
- Plug-and-play integration with any LLM provider (OpenAI, Anthropic, etc.)
- Deployable as a standalone API or embedded Python library
- Production-ready with customizable scanner pipelines
Use Cases
💡 Secure production LLM APIs against prompt injection attacks
💡 Anonymize PII in user inputs before sending to cloud LLMs
💡 Audit LLM outputs for toxic, biased, or non-factual content
💡 Prevent data leakage by scanning for secrets and sensitive information
💡 Build compliance-ready AI pipelines with content safety guardrails
Categories
Quick Start
```bash
pip install llm-guard
```
```python
from llm_guard.input_scanners import PromptInjection, Toxicity
from llm_guard.output_scanners import Bias, Relevance
# Scan user prompt
scanners = [PromptInjection(), Toxicity()]
sanitized_prompt, results, scores = scan_prompt(scanners, user_input)
# Scan LLM output
output_scanners = [Bias(), Relevance()]
sanitized_output, results, scores = scan_output(output_scanners, user_input, llm_output)
```