@squidcloud/client
    Preparing search index...

    Interface AiKnowledgeBaseGraphConfig

    Per-knowledge-base GraphRAG configuration. Opt-in and mutable (unlike embeddingModel/vectorDbType); only honored for vectorDbType:'mongoAtlas' KBs. Enabling it on a KB that already has content backfills the graph automatically (a debounced structural build runs once the KB goes quiet); the rebuildGraph endpoint remains available for an immediate or full rebuild.

    interface AiKnowledgeBaseGraphConfig {
        autoBuildDebounceMs?: number;
        autoBuildPaused?: boolean;
        concepts?: AiKnowledgeBaseGraphConceptsConfig;
        enabled: boolean;
        entityTypes?: string[];
        extractionModel?: AiChatModelSelection;
    }
    Index
    autoBuildDebounceMs?: number

    Quiet window (milliseconds) after the last graph activity before the sweep auto-builds the structure. Must be a positive integer, and is floored at 3 seconds — a shorter window is not servable. Defaults to the server's window (5 minutes); the sweep cadence (1 minute) bounds how promptly an elapsed window is picked up.

    autoBuildPaused?: boolean

    When true, the sweep does not start automatic structural builds for this KB: activity keeps arming the build, and the first quiet window after the flag is cleared runs it. Meant for an ingestion campaign whose builds an operator wants to schedule by hand; rebuildGraph stays available. Defaults to false.

    Concept-layer (facet taxonomy) configuration. Omitted ⇒ facets: 'auto'.

    enabled: boolean

    When true, ingestion extracts an entity/relationship graph and searchMode:'graph' becomes available.

    entityTypes?: string[]

    Optional domain taxonomy hint (entity types) injected into the extraction prompt.

    extractionModel?: AiChatModelSelection

    Chat model used for per-chunk entity/relationship extraction. Defaults to the server's graph model.