You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{name: 'PII Detection (Presidio)',category: 'pii',method: 'Algorithmic',description: 'Uses Microsoft Presidio for high-accuracy Named Entity Recognition (NER) to find and redact a wide range of PII entities.',link: 'https://github.com/microsoft/presidio'},
359
-
{name: 'Secrets Detection',category: 'security',method: 'Algorithmic',description: 'Scans text for patterns that match common API key and secret formats to prevent accidental leakage.',link: 'https://hub.guardrailsai.com/validator/guardrails/detect_secrets'},
360
-
{name: 'SQL Injection Detection',category: 'security',method: 'Algorithmic',description: 'Uses regular expressions to detect classic SQL injection attack patterns in user prompts.',link: 'https://hub.guardrailsai.com/validator/guardrails/sql_injection'},
359
+
{name: 'Secrets Present',category: 'security',method: 'Algorithmic',description: 'Scans text for patterns that match common API key and secret formats to prevent accidental leakage.',link: 'https://hub.guardrailsai.com/validator/guardrails/secrets_present'},
361
360
{name: 'Toxicity & Harmful Language',category: 'content',method: 'Model-Based',description: 'Uses a sentence-transformer model to classify text for toxicity, insults, threats, and other harmful content.',modelLink: 'https://huggingface.co/martin-ha/toxic-comment-model',link: 'https://hub.guardrailsai.com/validator/guardrails/toxic_language'},
362
361
{name: 'Profanity Filter',category: 'content',method: 'Algorithmic',description: 'A straightforward filter that checks for words against a comprehensive list of profanities.',link: 'https://hub.guardrailsai.com/validator/guardrails/profanity_free'},
363
-
{name: 'Detect Anonymized Text',category: 'pii',method: 'Model-Based',description: 'A specialized model that detects if a text has been *improperly* anonymized, finding leftover PII.',modelLink: 'https://huggingface.co/guardrails-ai/deberta-base-detection-v1',link: 'https://hub.guardrailsai.com/validator/guardrails/detect_anonymized_text'},
364
-
{name: 'Is High Quality',category: 'quality',method: 'Model-Based',description: 'Uses a model to score the grammatical and structural quality of a text, ensuring LLM outputs are well-formed.',modelLink: 'https://huggingface.co/guardrails-ai/distilbert-base-uncased-is-high-quality',link: 'https://hub.guardrailsai.com/validator/guardrails/is_high_quality'},
365
362
{name: 'Prompt Injection (Vigil)',category: 'security',method: 'Model-Based',description: 'A powerful prompt injection detector using a fine-tuned transformer model to identify adversarial inputs.',modelLink: 'https://huggingface.co/deadbits/vigil-llm-prompt-injection',link: 'https://github.com/deadbits/vigil-llm'},
366
-
{name: 'Prompt Injection (Heuristic)',category: 'security',method: 'Algorithmic',description: 'Detects prompt injection attempts by searching for common heuristic patterns and keywords (e.g., "ignore your instructions").',link: 'https://github.com/rebuff-ai/rebuff'},
367
-
{name: 'PII Redaction (Faker)',category: 'pii',method: 'Algorithmic',description: 'Detects PII and replaces it with realistic but fake data, which can be useful for testing and anonymization.',link: 'https://hub.guardrailsai.com/validator/guardrails/pii_redaction'},
368
363
{name: 'Gibberish Detection',category: 'quality',method: 'Algorithmic',description: 'Checks if text is coherent language or random characters by analyzing character frequencies and patterns.',link: 'https://hub.guardrailsai.com/validator/guardrails/gibberish_text'},
369
-
{name: 'Sentiment Analysis',category: 'content',method: 'Model-Based',description: 'Determines the sentiment (positive, negative, neutral) of a text to enforce a desired tone.',modelLink: 'https://huggingface.co/cardiffnlp/twitter-roberta-base-sentiment-latest',link: 'https://hub.guardrailsai.com/validator/guardrails/sentiment'},
370
-
{name: 'URL Reachability',category: 'quality',method: 'Algorithmic',description: 'Extracts all URLs from a response and verifies that they are reachable, preventing dead links.',link: 'https://hub.guardrailsai.com/validator/guardrails/urls_reachable'},
364
+
{name: 'URL Reachability',category: 'quality',method: 'Algorithmic',description: 'Extracts all URLs from a response and verifies that they are reachable, preventing dead links.',link: 'https://hub.guardrailsai.com/validator/guardrails/valid_url'},
371
365
{name: 'Code Security (Semgrep)',category: 'security',method: 'Algorithmic',description: 'Uses the Semgrep static analysis engine to scan for common security vulnerabilities in generated code.',link: 'https://github.com/semgrep/semgrep'},
372
-
{name: 'Language Detection',category: 'content',method: 'Algorithmic',description: 'Identifies the language of the text to enforce that the LLM only responds in specific languages.',link: 'https://hub.guardrailsai.com/validator/guardrails/detect_language'},
373
-
{name: 'Is Profane',category: 'content',method: 'Model-Based',description: 'A nuanced, model-based approach to detecting profanity that can understand context better than a simple word list.',modelLink: 'https://huggingface.co/guardrails-ai/distilbert-base-uncased-is-profane',link: 'https://hub.guardrailsai.com/validator/guardrails/is_profane'},
374
-
{name: 'Credit Card Number',category: 'pii',method: 'Algorithmic',description: 'Specifically detects and validates credit card numbers using regex and Luhn algorithm checks.',link: 'https://hub.guardrailsai.com/validator/guardrails/debit_card_number'},
375
-
{name: 'Email Address Validator',category: 'pii',method: 'Algorithmic',description: 'Checks if a string is a validly formatted email address.',link: 'https://hub.guardrailsai.com/validator/guardrails/is_valid_email'},
376
-
{name: 'Bias Detection',category: 'content',method: 'Model-Based',description: 'Uses a fine-tuned model to detect various forms of social bias (e.g., gender, race) in text.',modelLink: 'https://huggingface.co/d4data/bias-detection-model',link: 'https://github.com/d4data/bias-detection'},
366
+
{name: 'Language Detection',category: 'content',method: 'Algorithmic',description: 'Identifies the language of the text to enforce that the LLM only responds in specific languages.',link: 'https://hub.guardrailsai.com/validator/scb-10x/correct_language'},
377
367
{name: 'Emotion Detection',category: 'content',method: 'Model-Based',description: 'Classifies text into emotional categories like joy, sadness, anger, etc. Useful for monitoring conversation tone.',modelLink: 'https://huggingface.co/j-hartmann/emotion-english-distilroberta-base',link: 'https://huggingface.co/j-hartmann/emotion-english-distilroberta-base'},
378
-
{name: 'JSON Schema Validator',category: 'quality',method: 'Algorithmic',description: 'Ensures that an LLM\'s output strictly adheres to a provided JSON schema, crucial for reliable tool use.',link: 'https://hub.guardrailsai.com/validator/guardrails/is_json'},
379
-
{name: 'Word Count',category: 'quality',method: 'Algorithmic',description: 'Validates that the number of words in a text falls within a specified min/max range.',link: 'https://hub.guardrailsai.com/validator/guardrails/word_count'},
380
368
{name: 'Reading Time',category: 'quality',method: 'Algorithmic',description: 'Calculates the estimated reading time of a text and validates it against a min/max range.',link: 'https://hub.guardrailsai.com/validator/guardrails/reading_time'},
381
369
{name: 'Invisible Characters',category: 'security',method: 'Algorithmic',description: 'Detects and removes invisible characters from prompts that can be used to bypass other filters (from LLM Guard).',link: 'https://github.com/lakera/llm_guard'},
382
370
{name: 'On-Topic Check',category: 'content',method: 'Model-Based',description: 'Determines if a text is relevant to a list of predefined valid topics using semantic similarity.',modelLink: 'https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2',link: 'https://hub.guardrailsai.com/validator/guardrails/on_topic'},
{name: 'Valid URL',category: 'quality',method: 'Algorithmic',description: 'Checks if a string is a validly formatted URL.',link: 'https://hub.guardrailsai.com/validator/guardrails/is_valid_url'},
392
380
{name: 'Valid Range',category: 'quality',method: 'Algorithmic',description: 'Ensures a numerical value falls within a specified minimum and maximum range.',link: 'https://hub.guardrailsai.com/validator/guardrails/valid_range'},
393
381
{name: 'One Line',category: 'quality',method: 'Algorithmic',description: 'Validates that a string does not contain any newline characters.',link: 'https://hub.guardrailsai.com/validator/guardrails/one_line'},
382
+
{name: 'Valid JSON',category: 'quality',method: 'Algorithmic',description: 'Ensures that an LLM\'s output strictly adheres to a provided JSON schema, crucial for reliable tool use.',link: 'https://hub.guardrailsai.com/validator/guardrails/valid_json'},
394
383
],
395
384
gatewayComparison: {
396
385
recommendation: `Given the context of a software company comfortable with building and managing its own systems, **LiteLLM emerges as the more strategically aligned choice.** The "infrastructure-as-code" approach, deep extensibility via callbacks, and composable nature fit better with a skilled engineering team's desire for control and flexibility. While it requires more initial setup, it provides a more powerful and unopinionated foundation for a truly custom AI Control Plane.`,
0 commit comments