Content Filter

Filter blocked words and phrases from LLM message content.

Replaces occurrences of configured blocked words with ‘[FILTERED]’ in message content. Supports both string content and multimodal list content.

Configuration

Field	Type	Default	Description
blocked_words	array	[]	List of words or phrases to filter from content.

Examples

# basic
type: content_filter
config:
  blocked_words:
  - badword
  - secret