Graph - Verglas

Graph is a named Durable Object for property-graph data. Nodes and edges have stable IDs, a kind, and optional JSON properties. Edges are directed and may have a numeric weight.

Bind a graph

{
  "graphs": [
    { "binding": "KNOWLEDGE", "graph_name": "company-knowledge" }
  ]
}

Write nodes and edges

await env.KNOWLEDGE.upsertNodes([
  { id: "alice", kind: "person", properties: { team: "data" } },
  { id: "worker-docs", kind: "document", properties: { public: true } },
]);

await env.KNOWLEDGE.upsertEdges([
  {
    id: "alice-authored-worker-docs",
    from: "alice",
    to: "worker-docs",
    kind: "authored",
    weight: 1,
  },
]);

Traverse

const neighborhood = await env.KNOWLEDGE.neighbors("alice", {
  direction: "out",
  edgeKinds: ["authored"],
  depth: 2,
  limit: 100,
  returnNodes: true,
  returnEdges: true,
});

const path = await env.KNOWLEDGE.shortestPath("alice", "worker-docs", {
  direction: "both",
  maxDepth: 4,
});

Depth is bounded from 1 through 8. Traversal limits are part of the API so one request cannot turn into an unbounded graph scan.

Complete API

Method	Purpose
`upsertNodes`, `upsertEdges`	Add or fully replace entities
`getNodes`, `getEdges`	Fetch entities in requested ID order
`deleteNodes`	Delete nodes and incident edges atomically
`deleteEdges`	Delete edges by ID
`neighbors`	Traverse a bounded neighborhood
`shortestPath`	Find one deterministic unweighted path
`describe`	Inspect graph identity and durable counts

Create property indexes for fields used in nodeFilter or edgeFilter. A Graph is best for bounded application traversal; publish event history to Iceberg when you also need large analytical scans.