Skip to main content
Nodes are the building blocks of your agent graph. You add them to the canvas, connect edges, and run the graph to move data from one node to the next. A single node on the agent composer canvas

The Node Block (on the Canvas)

Each node appears as a block on the canvas with:
  • ID: The label shown on the node (e.g., Agent_2).
  • Inputs (left side): The data this node consumes.
  • Outputs (right side): The data this node produces.
  • Config Ports (bottom): Optional connectors for configurations used by the node (varies by node type).

Inputs

Inputs are how data enters a node.
  • Inputs are named ports (e.g., input) shown on the left side of the node.
  • Each input receives a list of items from upstream nodes.
  • Some nodes (like triggers) can run without inputs; others require them to execute.

Outputs

Outputs are the node’s results.
  • Outputs are named ports shown on the right side of the node.
  • Each output emits a list of items that can feed multiple downstream nodes.
  • Every node must produce every output it declares (so your graph wiring stays consistent).

Parameters (Node Settings)

Parameters control a node’s behavior and are configured in the node details panel.
  • Parameters are typed and validated values.
  • Required parameters must be filled; optional ones use defaults when provided.
  • Parameters can be plain values or expressions that reference upstream inputs/outputs. Expressions allow dynamic behavior based on runtime data. For example, you can set a parameter to {input} to use the node’s input data or {@NodeA.output} to reference the output of an upstream node. Read more about using Expressions in parameters.
Node inspector for an Agent node

Config Nodes (Optional)

Config nodes provide static configuration that other nodes can consume.
  • A node can declare config ports that accept config nodes.
  • Config nodes run before the main node so their outputs are available at run time.
  • Use config nodes for reusable settings (models, tools, credentials, shared options).

Node List (Node Picker)

The node picker is how you add nodes to the canvas.
  • Search to find nodes quickly.
  • Categories group nodes by purpose:
  • Agent – agents with prompts, tools, memory.
  • Trigger – starting points for the graph.
  • Utility – helpers, routing, parsing, transforms.
  • Model – model configs.
  • Tool – external tool connectors.
Node inspector for an Agent node