How to read a node page
Every page in the reference has the same shape and is built from the node registry — the same one the editor builds its palette from. So there is never a field the app does not have, and never a port drawn from memory.
The card
At the top is the node exactly as it appears on the canvas: icon, name, and port dots on the left and right.
The shape of a dot says what travels through it:
- circle — data only;
- triangle — execution only (order, no value);
- diamond — both, on one edge.
The colour is the wire type, in the same colours the canvas uses. The full list is on the Wires page.
Badges
Next to the name you may see:
- Tool — the node can be attached to an agent as a tool, and the model calls it itself;
- Exec — the node has an execution input, so it is a step in the graph rather than a source of configuration;
- Subgraph — other nodes run inside it;
- Stateful — it keeps state between runs.
Ports
Two tables, inputs and outputs. Two of their columns are not on the card:
- Payload — the semantic type of the value (
message,dataframe,file…); - Notes — whether the port is required, whether it takes several edges, and under which setting it appears at all. A conditional port may be missing from the card: it shows up once you enable the matching option.
Configuration
The field table follows the same order and grouping as the form in the editor: primary fields first, then Advanced. Each field shows its type, its default, the options for enumerations, and markers — whether it takes templates, whether a connected port overrides it, whether it is a secret.
Three fields (expose_error_output, split_ports_in, split_ports_out) are folded into
their own block: every node has them, added by the platform.
The example
Almost every page carries a minimal workflow with a Copy nodes button. Press it, open the editor and hit Ctrl+V on the canvas — the whole graph is pasted, wires included. A pasted entry maps onto the one already on the canvas, so pasting into a non-empty graph is fine.