Agentic Query API

Endpoint

POST /graphs/{graphId}/agentic-query

Request body

{
  "query": "Who works at technology companies connected to Alice?",
  "topK": 5,
  "depth": 2,
  "direction": "outbound",
  "nodeType": "Person"
}

Field	Type	Required	Default	Description
query	string	yes	—	Natural language or semantic text
topK	number	no	5	Number of vector similarity results
depth	number	no	2	Traversal depth from each hit
direction	string	no	outbound	Edge direction for traversal
nodeType	string	no	all	Restrict vector search to a node type

Response

{
  "query": "Who works at technology companies connected to Alice?",
  "vectorHits": [
    { "nodeId": "node_alice", "score": 0.94 }
  ],
  "context": {
    "nodes": [ ... ],
    "edges": [ ... ]
  },
  "tokenEstimate": 420
}

tokenEstimate is a rough count of the serialised context size in tokens — useful for LLM prompt budget planning.

Using the context in a prompt

const result = await client.agenticQuery(graphId, { query });

const systemPrompt = [
  "You are a helpful assistant.",
  "Use the following graph context to answer the user's question.",
  "Graph nodes: " + JSON.stringify(result.context.nodes),
  "Graph edges: " + JSON.stringify(result.context.edges),
].join("\n");