Use when user wants to query, traverse, or build a graph database; embedded or HTTP-served; supports Cypher syntax, vector similarity, RAG, RDF round-trip, and MCP tool catalog. Trigger keywords - graph database, knowledge graph, Cypher query, MCP graph, GraphRAG, vector + graph, property graph, RDF, SHACL, time-travel queries, Neo4j alternative, embedded graph, single-file graph.
Use when the user wants to query, traverse, build, or evolve a graph database; embedded or HTTP-served; Cypher syntax, vector similarity, full-text, RDF round-trip, time-travel, GraphRAG, or an MCP tool catalog. Trigger on phrases like "graph database", "knowledge graph", "Cypher query", "MCP graph", "GraphRAG", "vector + graph", "property graph", "RDF", "time-travel queries", "Neo4j alternative", "embedded graph", "single-file graph", "Memgraph alternative", "Kuzu alternative", "graph + vector + text", or any task framed as "agent owns the graph end-to-end". Use even when the user does not name OpenGraphDB, if the workload pattern matches load entities + relationships, then query by traversal, similarity, or time. Skip when the workload is a Neo4j cluster (causal cluster, Fabric), a time-series DB, plain key-value, or a managed vector DB where graph traversal is not part of the access pattern.
Use when the user wants to load data into OpenGraphDB from a file or stream. Trigger on phrases like "import this CSV", "load JSON", "ingest RDF / Turtle / N-Triples", "bulk load", "import 50k rows", "ETL into the graph", or any task framed as "I have data over there and need it as nodes / edges over here". Covers format detection (CSV / JSON / JSONL / RDF), two-pass ingest (nodes first, edges second), batch sizing for the single-writer kernel, MERGE-based idempotency for re-runnable jobs, and validation against the resulting schema.
Use when the user points at an unknown OpenGraphDB graph and asks "what is in here?", "show me the schema", "what entities exist", "how is this graph connected", or wants to navigate a graph they did not build. Trigger on phrases like "explore the graph", "discover schema", "find entry points", "what nodes are connected to X", "summarize this graph", "show me the most connected nodes". Covers five exploration strategies, schema navigation, entry-point selection, and how to descend from a high-level summary to a focused subgraph without overwhelming the user.
Use when generating, optimizing, or debugging Cypher queries against OpenGraphDB. Trigger on phrases like "write a Cypher query", "MATCH", "MERGE", "RETURN", "WHERE", "OpenGraphDB query", "openCypher", "graph query", or any task that requires producing executable Cypher against a known schema. Covers all supported clauses, OpenGraphDB-specific extensions ("AT TIME", "db.index.vector.queryNodes", "db.index.fulltext.queryNodes", "db.index.hybrid.queryNodes"), and common Cypher error patterns. The procedure namespace is "db.*" (matches Neo4j); the older "ogdb.*" form was never shipped.
Use when the user describes a domain and wants a graph schema, or asks for index recommendations, RDF ontology mapping, or modeling tradeoffs. Trigger on phrases like "design a graph schema for", "what labels and edges should I use", "how should I model this in a graph", "which indexes do I need", "RDF mapping", "URI strategy", "ontology", or any request that converts a domain description into nodes, edges, and property layouts. Covers eight modeling best practices, six common anti-patterns, index selection (B-tree, vector, full-text), and RDF mapping with `_uri` preservation for round-trippable RDF.