Semantic Index Configuration form

  • Release version: Australia
  • Updated July 6, 2026
  • 3 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Semantic Index Configuration form

    The Semantic Index Configuration form in ServiceNow enables you to define semantic indexing settings for an AI Search indexed source. This configuration affects how content is semantically indexed to enhance search relevance and results. The form is available only if the AI Search Semantic Controller plugin (com.glide.ais.semanticsearch) is active, which requires at least one Now Assist application installed on your instance.

    Show full answer Show less

    Key Features

    • Name: Assign a unique, special-character-free name to the semantic index generated by this configuration. For example, "Knowledge-Table-semantic-index" for a Knowledge Table source.
    • Embedding Models: Select from predefined embedding models used to encode the indexed content semantically:
      • ServiceNow Embedding (E5) – default model with a 512-term encoder limit.
      • Azure OpenAI Embedding – external fine-tuned model.
      • Google Gemini Embedding – external fine-tuned model.
      • Custom Embedding – user-defined fine-tuned model.
    • Active: Toggle to activate or deactivate the semantic index configuration. Inactive configurations are ignored during indexing.
    • Indexed Source: Automatically set reference to the AI Search indexed source this configuration applies to.
    • Application: Automatically set application scope for the configuration record.
    • Chunking Configuration for Embedding: Controls how text is split (chunked) for embedding:
      • Chunking Strategy: Choose among
        • Passage: Suitable for longer text; splits text into chunks based on a chunk unit and size.
        • Truncate: For short text; concatenates fields and indexes up to a maximum word limit.
        • Full Text: For short text; indexes all terms up to the embedding model’s encoder limit.
      • Overlap Sentences: (Passage strategy only) Number of sentences overlapping between chunks to increase search recall at some performance cost. Default is 5.
      • Chunk Unit: (Passage strategy only) Defines whether chunk size is measured in words or sentences.
      • Chunk Size: (Passage strategy only) Maximum number of words or sentences per chunk. Defaults are 250 words or 15 sentences.
      • Maximum Total Words: (Truncate strategy only) Maximum words indexed from concatenated semantic fields, defaulting to 500 words.

    Practical Benefits for ServiceNow Customers

    By configuring semantic indexing with this form, you improve AI Search’s ability to understand and retrieve relevant content based on semantic meaning rather than simple keyword matching. Selecting appropriate embedding models and chunking strategies allows you to optimize indexing performance and accuracy based on your content type and search needs. Activating the configuration ensures indexing aligns with your business requirements, while the chunking options let you tailor how long or short texts are processed for embedding, enhancing search precision and recall.

    The Semantic Index Configuration form enables you to define semantic indexing settings for an AI Search indexed source.

    For details on defining and modifying semantic indexing settings for an indexed source, see Configure semantic indexing settings for an indexed source.
    Note:
    This form is only available when the AI Search Semantic Controller plugin (com.glide.ais.semantic_search) is active on your instance. To activate this plugin, your instance must have at least one Now Assist application installed.
    Table 1. Semantic Index Configuration form
    Field Description
    Name Unique name for the semantic index generated by this semantic index configuration. As an example, if you're creating a semantic index configuration for the Knowledge Table indexed source, you might name it Knowledge-Table-semantic-index.
    Note:
    The semantic index's name can't contain special characters, underscores, or whitespace.
    Embedding Models List of embedding models to use for the semantic index configuration.
    • Default value: ServiceNow Embedding (E5)
    • Supported values:
      • ServiceNow Embedding (E5): Use the E5 fine-tuned embedding model for content in the semantic index. The embedding model's encoder limit is 512 terms.
      • Azure OpenAI Embedding: Use the Azure OpenAI fine-tuned embedding model for content in the semantic index. For more information, see Configuring an external or custom embedding model.
      • Google Gemini Embedding: Use the Google Gemini fine-tuned embedding model for content in the semantic index. For more information, see Configuring an external or custom embedding model.
      • Custom Embedding: Use the custom fine-tuned embedding model for content in the semantic index. For more information, see ../concept/creating-byom.html.
    Active Option to make the semantic index configuration active for your instance. AI Search ignores inactive semantic index configurations when indexing content from the specified index source.
    Indexed Source Reference to the AI Search indexed source that you want to apply this semantic index configuration to. This field is automatically set.

    For more details on indexed sources, see Indexed sources in AI Search.
    Application Application scope for the semantic index configuration record. This field is automatically set.
    Chunking Configuration For Embedding
    Chunking Strategy Strategy to use when chunking semantically indexed text for the embedding model.
    • Default value: Passage
    • Supported values:
      • Passage: Chunking strategy for longer text field values. Index text from semantic field values in chunks. Each chunk contains a maximum number of words or sentences determined by your Chunk Unit and Chunk Size selections.
      • Truncate: Chunking strategy for short text field values. Concatenate all semantic index field values, then perform semantic indexing for terms up to the Maximum Total Words limit.
      • Full Text: Chunking strategy for short text field values. Concatenate all semantic index fields, then perform semantic indexing for all terms up to the embedding model's encoder limit.
    • Type: choice list
    Overlap Sentences Number of sentences to overlap between chunks when indexing text from semantic index field values. Higher overlap values increase recall for semantic vector search at the expense of performance.

    This field appears only when Passage is selected from Chunking Strategy.

    • Default value: 5
    • Supported values: Any non-negative integer
    • Type: integer
    Chunk Unit Textual unit to use as the basis for chunk size when indexing semantic field values for semantic vector search.
    This field appears only when Passage is selected from Chunking Strategy.
    • Default value: Words
    • Supported values:
      • Words: Use words as the textual unit by which semantic index field values are chunked. Each chunk can include up to Chunk Size words.
      • Sentences: Use sentences as the textual unit by which semantic index field values are chunked. Each chunk can include up to Chunk Size sentences.
    • Type: choice list
    Chunk Size Maximum number of words or sentences (depending on your Chunk Unit selection) to include in a chunk when indexing semantic field values for semantic vector search.
    This field appears only when Passage is selected from Chunking Strategy.
    • Default value: 250 when Words is selected from Chunk Unit, or 15 when Sentences is selected from Chunk Unit
    • Supported values: Any non-negative integer
    • Type: integer
    Maximum Total Words Maximum number of words to index for semantic vector search from the concatenated values of all semantic index fields.
    This field appears only when Truncate is selected from Chunking Strategy.
    • Default value: 500
    • Supported values: Any non-negative integer
    • Type: integer