| # graph_query | |
| Query the TigerGraph knowledge graph using natural language. Performs dual-level keyword extraction, entity vector search, and multi-hop graph traversal to find relevant entities, relationships, and evidence passages. | |
| ## Parameters | |
| - `query` (string, required): Natural language question to search the knowledge graph | |
| - `depth` (integer, optional, default=2): Number of hops for graph traversal (1-4) | |
| - `top_k` (integer, optional, default=5): Number of seed entities to retrieve | |
| ## Returns | |
| JSON object with: | |
| - `entities`: List of entities found with names, types, and descriptions | |
| - `relations`: List of relationships traversed with source, target, and type | |
| - `passages`: Relevant text chunks connected to discovered entities | |
| - `reasoning_path`: Step-by-step explanation of the graph traversal | |
| ## Example | |
| ``` | |
| graph_query "Were Scott Derrickson and Ed Wood of the same nationality?" --depth 2 | |
| ``` | |
| ## Notes | |
| - Requires TigerGraph connection (set TG_HOST, TG_PASSWORD env vars) | |
| - Falls back to in-memory entity extraction if TigerGraph unavailable | |
| - Uses the configured LLM provider for keyword extraction | |