Graphs

A Graph defines the deterministic structure of a workflow — which nodes exist, how they connect, and the conditions under which edges are traversed. They can be cyclic or acyclic depending on the need.

Creating a graph

Use the createGraph helper to build a validated Graph object. The id is auto-generated if omitted:

import { createGraph } from '@cycgraph/orchestrator';

const graph = createGraph({
  name: "Research Pipeline",
  description: "Searches the web and writes a summary",
  start_node: "researcher",
  end_nodes: ["writer"],
  nodes: [
    // ... Node definitions ...
  ],
  edges: [
    // ... Edge definitions ...
  ]
});

Graph configuration

Field	Type	Default	Description
id	string (UUID)	auto-generated	Unique identifier for the graph definition.
name	string	required	Human-readable name.
description	string	required	Description of what this graph does.
nodes	GraphNode[]	required	List of nodes that define the work to be done.
edges	GraphEdge[]	required	Directed edges defining the flow of execution.
start_node	string	required	The ID of the first node to execute.
end_nodes	string[]	required	Terminal node IDs. Execution stops here.

Edges

An Edge is a directed connection between a source node and a target node.

When a node completes, the orchestrator evaluates all outgoing edges from that node. The condition determines whether the edge is traversed.

Edge configuration

Field	Type	Default	Description
id	string (UUID)	auto-generated	Unique identifier for the edge (used in validation messages and debug logs).
source	string	required	Source node ID.
target	string	required	Target node ID.
condition	EdgeCondition	{ type: 'always' }	Routing logic.
metadata	object	{}	Arbitrary metadata for tooling/debugging.

Edge conditions

Type	Description	Required fields
always	Unconditional routing. The edge is always traversed.	—
conditional	Dynamic routing. Evaluates an expression against the workflow’s memory state.	condition: string
map	Specialized edge used exclusively by map-reduce fan-out nodes.	—

Conditional example:

Loops back to the writer if score is low.

{
  source: "evaluator",
  target: "writer",
  condition: {
    type: "conditional",
    condition: "memory.draft_score < 0.8"
  }
}

Next steps

• Nodes — node types, configuration, and state slicing
• Workflow State — the shared state object
• Agents — how agent nodes work