Implementation:Protectai Llm guard Input Scanner Base

Knowledge Sources	LLM Guard
Domains	Software_Design, Security, Plugin_Architecture
Last Updated	2026-02-14 12:00 GMT

Overview

Concrete abstract base class defining the input scanner interface that all LLM Guard input scanners must implement.

Description

The Scanner base class in llm_guard.input_scanners.base defines the interface contract for all input scanners. It requires implementing a single scan method that takes a prompt string and returns a tuple of (sanitized_prompt, is_valid, risk_score).

Usage

Subclass this base class when creating custom input scanners for LLM Guard.

Code Reference

Source Location

• Repository: llm-guard
• File: llm_guard/input_scanners/base.py

Signature

class Scanner:
    def scan(self, prompt: str) -> tuple[str, bool, float]:
        """
        Scan a prompt string.

        Args:
            prompt: The input text to scan.

        Returns:
            tuple: (sanitized_prompt, is_valid, risk_score)
                - sanitized_prompt: The processed prompt
                - is_valid: True if prompt passes the scanner
                - risk_score: Float in [0,1] or -1 if not applicable
        """
        raise NotImplementedError

Import

from llm_guard.input_scanners.base import Scanner

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	Input text to scan

Outputs

Name	Type	Description
sanitized_prompt	str	Processed prompt (may be modified by scanner)
is_valid	bool	True if prompt passes, False if fails
risk_score	float	Risk score [0,1] or -1 if not applicable

Usage Examples

Custom Scanner Implementation

from llm_guard.input_scanners.base import Scanner

class CustomKeywordScanner(Scanner):
    def __init__(self, banned_words: list[str]):
        self._banned_words = banned_words

    def scan(self, prompt: str) -> tuple[str, bool, float]:
        prompt_lower = prompt.lower()
        for word in self._banned_words:
            if word.lower() in prompt_lower:
                return prompt, False, 1.0
        return prompt, True, -1.0

# Use in pipeline
from llm_guard import scan_prompt
scanner = CustomKeywordScanner(["classified", "secret"])
sanitized, valid, scores = scan_prompt([scanner], "This is classified info")

Related Pages

Implements Principle

• Principle:Protectai_Llm_guard_Input_Scanner_Interface

Requires Environment

• Environment:Protectai_Llm_guard_Python_Runtime_Dependencies