Interface ClusterConfig

Clustering algorithm to use.

How points are colored in the renderer: by their assigned cluster (default) or by their source entity (useful for multi-entity analyses). Persisted so a reloaded analysis restores the chosen legend mode.

Number of dimensions for the projected layout (2 or 3). Defaults to 2.

Distance metric used when comparing vectors.

Specific Entity Document ID supplying the vectors. Blank => source adapter chooses.

Multi-entity clustering: one or more Entity Document IDs whose vectors are merged into a single analysis. When supplied (length >= 1) this takes precedence over EntityDocumentID. All selected documents MUST share the same embedding model + vector dimensionality — the source adapter hard-blocks incompatible combinations rather than produce a meaningless layout.

Entity ID — persisted to the top-level EntityID column when available. Null for multi-entity analyses.

Entity name whose records are being clustered (drives drilldown / view qualification).

Epsilon neighbourhood radius for DBSCAN (ignored for K-Means).

Optional SQL-style filter applied by the vector source.

Number of clusters for K-Means (ignored for DBSCAN).

Maximum number of records to fetch for clustering.

Minimum points to form a dense region in DBSCAN (ignored for K-Means).

When true, the engine runs the optional LLM cluster-naming step.