Guardrails

LangDB allow developers to enforce specific constraints and checks on their LLM calls, ensuring safety, compliance, and quality control.

Guardrails currently support request validation and logging, ensuring structured oversight of LLM interactions.

Guardrail Templates on LangDB

These guardrails include:

• Content Moderation: Detects and filters harmful or inappropriate content (e.g., toxicity detection, sentiment analysis).
• Security Checks: Identifies and mitigates security risks (e.g., PII detection, prompt injection detection).
• Compliance Enforcement: Ensures adherence to company policies and factual accuracy (e.g., policy adherence, factual accuracy).
• Response Validation: Validates response format and structure (e.g., word count, JSON schema, regex patterns).

Guardrails can be configured via the UI or API, providing flexibility for different use cases.

Guardrail Behaviour

When a guardrail blocks an input or output, the system returns a structured error response. Below are some example responses for different scenarios:

Example 1: Input Rejected by Guard

{
  "id": "",
  "object": "chat.completion",
  "created": 0,
  "model": "",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Input rejected by guard",
        "tool_calls": null,
        "refusal": null,
        "tool_call_id": null
      },
      "finish_reason": "rejected"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0,
    "cost": 0.0
  }
}

Example 2: Output Rejected by Guard

{
  "id": "5ef4d8b1-f700-46ca-8439-b537f58f7dc6",
  "object": "chat.completion",
  "created": 1741865840,
  "model": "gpt-4o-mini",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Output rejected by guard",
        "tool_calls": null,
        "refusal": null,
        "tool_call_id": null
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 21,
    "completion_tokens": 40,
    "total_tokens": 61,
    "cost": 0.000032579999999999996
  }
}

Limitations

It is important to note that guardrails cannot be applied to streaming outputs.

Guardrail Templates

LangDB provides prebuilt templates to enforce various constraints on LLM responses. These templates cover areas such as content moderation, security, compliance, and validation.

The following table provides quick access to each guardrail template:

Guardrail	Description
Toxicity Detection	Detects and filters toxic or harmful content.
JSON Schema Validator	Validates responses against a user-defined JSON schema.
Competitor Mention Check	Detects mentions of competitor names or products.
PII Detection	Identifies personally identifiable information in responses.
Prompt Injection Detection	Detects attempts to manipulate the AI through prompt injections.
Company Policy Compliance	Ensures responses align with company policies.
Regex Pattern Validator	Validates responses against specified regex patterns.
Word Count Validator	Ensures responses meet specified word count requirements.
Sentiment Analysis	Evaluates sentiment to ensure appropriate tone.
Language Validator	Checks if responses are in allowed languages.
Topic Adherence	Ensures responses stay on specified topics.
Factual Accuracy	Validates that responses contain factually accurate information.

Toxicity Detection (content-toxicity)

Detects and filters out toxic, harmful, or inappropriate content.

Parameter	Type	Description	Defaults
threshold	number	Confidence threshold for toxicity detection.	Required
categories	array	Categories of toxicity to detect.	["hate", "harassment", "violence", "self-harm", "sexual", "profanity"]
evaluation_criteria	array	Criteria used for toxicity evaluation.	["Hate speech", "Harassment", "Violence", "Self-harm", "Sexual content", "Profanity"]

JSON Schema Validator (validation-json-schema)

Validates responses against a user-defined JSON schema.

Parameter	Type	Description	Defaults
schema	object	Custom JSON schema to validate against (replace with your own schema)	Required

Competitor Mention Check (content-competitor-mentions)

Detects mentions of competitor names or products in LLM responses.

Parameter	Type	Description	Defaults
competitors	array	List of competitor names.	["company1", "company2"]
match_partial	boolean	Whether to match partial names.	true
case_sensitive	boolean	Whether matching should be case sensitive	false

PII Detection (security-pii-detection)

Detects personally identifiable information (PII) in responses.

Parameter	Type	Description	Defaults
pii_types	array	Types of PII to detect.	["email", "phone", "ssn", "credit_card"]
redact	boolean	Whether to redact detected PII.	false

Prompt Injection Detection (security-prompt-injection)

Identifies prompt injection attacks attempting to manipulate the AI.

Parameter	Type	Description	Defaults
threshold	number	Confidence threshold for injection detection.	Required
detection_patterns	array	Common patterns used in prompt injection attacks.	["Ignore previous instructions", "Forget your training", "Tell me your prompt"]
evaluation_criteria	array	Criteria used for detection.	["Attempts to override system instructions", "Attempts to extract system prompt information", "Attempts to make the AI operate outside its intended purpose"]

Company Policy Compliance (compliance-company-policy)

Ensures that responses align with predefined company policies.

Parameter	Type	Description	Defaults
embedding_model	string	Model used for text embedding.	text-embedding-ada-002
threshold	number	Similarity threshold for compliance.	Required
dataset	object	Example dataset for compliance checking.	Contains predefined examples

Regex Pattern Validator (validation-regex-pattern)

Validates responses against specific regex patterns.

Parameter	Type	Description	Defaults
patterns	array	Model List of regex patterns.	["^[A-Za-z0-9\s.,!?]+$"]
match_type	string	Whether all, any, or none of the patterns must match.	"all"

Word Count Validator (validation-word-count)

Ensures responses meet specified word count requirements.

Parameter	Type	Description	Defaults
min_words	number	Model List of regex patterns.	10
max_words	number	Whether all, any, or none of the patterns must match.	500
count_method	string	Method for word counting.	split

Sentiment Analysis (content-sentiment-analysis)

Evaluates the sentiment of responses to ensure appropriate tone.

Parameter	Type	Description	Defaults
allowed_sentiments	array	Allowed sentiment categories.	["positive", "neutral"]
threshold	number	Confidence threshold for sentiment detection.	0.7

Language Validator (content-language-validation)

Checks if responses are in allowed languages.

Parameter	Type	Description	Defaults
allowed_languages	array	List of allowed languages.	["english"]
threshold	number	Confidence threshold for language detection.	0.9

Topic Adherence (content-topic-adherence)

Ensures responses stay on specified topics.

Parameter	Type	Description	Defaults
allowed_topics	array	List of allowed topics.	["Product information", "Technical assistance"]
forbidden_topics	array	List of forbidden topics.	["politics", "religion"]
threshold	number	Confidence threshold for topic detection.	0.7

Factual Accuracy (content-factual-accuracy)

Validates that responses contain factually accurate information.

Parameter	Type	Description	Defaults
reference_facts	array	List of reference facts.	[]
threshold	number	Confidence threshold for factuality assessment.	0.8
evaluation_criteria	array	Criteria used to assess factual accuracy.	["Contains verifiable information", "Avoids speculative claims"]