Member Junction
    Preparing search index...

    Interface EntityDocumentConfiguration

    Configuration interface for EntityDocument.Configuration JSON column.

    Controls how the vectorization pipeline processes records for a given entity document — which fields get into vector metadata, how they're truncated, how records are filtered, and pipeline tuning knobs.

    All fields are optional. NULL / missing means "use system defaults."

    interface EntityDocumentConfiguration {
        chunking?: EntityDocumentChunkingConfig;
        metadata?: EntityDocumentMetadataConfig;
        pipeline?: EntityDocumentPipelineConfig;
        recordFilter?: EntityDocumentRecordFilterConfig;
        scheduling?: EntityDocumentSchedulingConfig;
        vectorIdStrategy?: "hash" | "recordId";
    }
    Index

    Properties

    chunking? metadata? pipeline? recordFilter? scheduling? vectorIdStrategy?

    Properties

    chunking?: EntityDocumentChunkingConfig

    Content chunking for long-form text fields

    metadata?: EntityDocumentMetadataConfig

    Controls which entity fields appear in vector metadata and how they're stored

    pipeline?: EntityDocumentPipelineConfig

    Pipeline tuning — batch sizes, concurrency, rate limiting

    recordFilter?: EntityDocumentRecordFilterConfig

    Record filtering — limit which records get vectorized

    scheduling?: EntityDocumentSchedulingConfig

    Scheduling settings for automated incremental sync

    vectorIdStrategy?: "hash" | "recordId"

    Strategy for generating vector record IDs in the vector database:

    • "hash" (default): SHA-1 hash of <EntityDocumentID>_<compositeKey>. Safe everywhere — stays under provider ID length limits, handles composite primary keys, and namespaces by entity document so two documents writing the same entity into one index never collide.
    • "recordId": the source record's primary key value is used directly as the vector ID (UUIDs are normalized to lowercase for cross-platform portability; composite PK values are joined with "||"). Makes vectors portable between systems and lets the record be identified without a RecordID metadata field. Requires that only ONE entity document writes to the index (or namespace) — otherwise IDs collide. Changing this on an already-populated index orphans previously-written vectors under their old hashed IDs; purge or re-create the index when switching.

    Settings

    Member Visibility

    On This Page

    Properties