Package org.springframework.ai.converter

Class ThinkingTagCleaner

java.lang.Object
org.springframework.ai.converter.ThinkingTagCleaner
All Implemented Interfaces:
ResponseTextCleaner
public class ThinkingTagCleaner extends Object implements ResponseTextCleaner
A ResponseTextCleaner that removes thinking tags from LLM responses. This cleaner supports multiple tag patterns to handle different AI models:
  • Amazon Nova: <thinking>...</thinking>
  • Qwen models: <think>...</think>
  • DeepSeek models: various thinking patterns
  • Claude models: thinking blocks in different formats

Performance: This cleaner includes fast-path optimization. For responses without thinking tags (most models), it performs a quick character check and returns immediately, making it safe to use as a default cleaner even for non-thinking models.

Since:
1.1.0
Author:
liugddx

  • Constructor Details

    • ThinkingTagCleaner

      public ThinkingTagCleaner()
      Creates a cleaner with default thinking tag patterns.

    • ThinkingTagCleaner

      public ThinkingTagCleaner(List<Pattern> patterns)
      Creates a cleaner with custom patterns.
      Parameters:
      patterns - the list of regex patterns to match thinking tags

    • ThinkingTagCleaner

      public ThinkingTagCleaner(String... patternStrings)
      Creates a cleaner with custom pattern strings.
      Parameters:
      patternStrings - the list of regex pattern strings to match thinking tags

  • Method Details

    • clean

      public String clean(String text)
      Description copied from interface: ResponseTextCleaner
      Clean the given text by removing unwanted patterns, tags, or formatting.
      Specified by:
      clean in interface ResponseTextCleaner
      Parameters:
      text - the raw text from LLM response
      Returns:
      the cleaned text ready for parsing

    • builder

      public static ThinkingTagCleaner.Builder builder()
      Creates a builder for constructing a thinking tag cleaner.
      Returns:
      a new builder instance