Edge types

Edges connect nodes. Most edges carry data flow and drive the execution schedule (the topological order). Two edge types are non-flow: they’re read or followed by special-case logic and are deliberately excluded from scheduling.

To set an edge’s type, draw a connection on the canvas and pick the type in the edge menu.

Flow edges

These participate in topological scheduling — a node runs once all its incoming flow edges are satisfied.

Type	Behavior
Data (default)	The source’s output is passed to the target as input. The standard connection.
Event	Pure signaling — the target runs after the source completes, but no data is passed.
Control	Explicit gating: the target waits for the source. Used by Condition nodes to gate a branch.
Success	Followed only when the source node succeeds.

Non-flow edges

These are excluded from the topological sort and from a node’s predecessors/successors, so they never block or trigger scheduling. They’re handled by dedicated logic instead.

Type	Behavior
Fail	Followed only when the source node fails. May point backward to form a retry loop. Drives error handling, not the DAG.
Membership	A structural reference: a coordinator (Panel / Orchestrator) points at the Agent reference nodes it convenes. Read off the graph to build the roster — it carries no data and imposes no ordering.

Caution

Because Fail and Membership edges are excluded from scheduling, you can connect a Panel to its member agents (Membership) without creating a data dependency or a cycle. The members are convened inside the Panel’s execution, not scheduled before it.

Edge fields

Every edge stores:

• source, target — the connected node ids.
• edge_type — one of the six types above (data is the default).
• condition_branch — the branch label a Condition node uses to choose this edge.
• label — an optional display label.

The canonical definition lives in backend/src/graph/schema.rs (EdgeType), and the scheduling filter (EdgeType::is_flow) is what excludes Fail and Membership from the topological sort.