azure-kusto-graph

Use the edges-first pattern: define an edges table with source and target columns, define node lookup tables with IDs and properties, union them, then pipe edges into make-graph Source --> Target with nodes on nodeId. The result can be queried with graph-match or visualized in Kusto Explorer.

Use graph-shortest-paths with a variable-length edge pattern such as (start)-[e*1..10]->(end) after building a graph with make-graph. Filter start and end nodes in the where clause and project array_length(e) for the path length.

Transient graphs are built inline with make-graph during query execution and suit ad-hoc analysis. Persistent graphs use graph models and snapshots stored in database metadata, offering enterprise scale and reuse across queries but requiring setup commands.

Yes. End the query at make-graph without piping to graph-match, and Kusto Explorer opens its interactive graph visualization window. For ADX Web Explorer, append graph-to-table nodes as N, edges as E since it cannot render graph objects.

No. It expects a working KQL query or known table as input plus a natural-language description of the desired graph structure. For general natural-language-to-KQL conversion, a dedicated query-generation skill should be used instead.

By default graph-match uses cycles = unique_edges, which can still produce repeated node visits. Set cycles = none to prevent node revisits, and add distinct or explicit node inequality constraints like u1.nodeId != u2.nodeId for bidirectional patterns.