Skip to main content

Nodes reference

This page describes all node types and attachment tabs available in the workflow builder.

The builder panel has two distinct sections:

  • Components — node types you drag onto the canvas to build the execution graph (Agent, Workflow, Orchestrator)
  • Attachment tabs — items you drag onto an existing canvas node to configure it (Sources, Tools, Skills, LLMs)

Two workflow contexts​

Standalone and assistant-embedded workflows use the same canvas but have different builder panels:

FeatureStandalone workflowAssistant workflow
Access path/workflows-task/\{id\}/assistants-config/\{id\}/workflow/\{id\}
Components panel (Agent / Workflow / Orchestrator)YesNo
Node Mode selector (Agent ↔ Orchestrator)NoYes
Attachment tabs (Sources / Tools / Skills / LLMs)YesYes
Run / History / ScheduleYesNo

Canvas nodes​

Input​

The Input node is always present on the canvas and cannot be deleted.

Configure here:

  • Input parameters — fields users fill in at run time (see Input parameters)
  • Prompt template — shared text snippet referenceable in component nodes via {{variable}} placeholders

The Input node is the starting point of every execution path.

Components​

Standalone workflow Components panel showing Agent, Workflow, and Orchestrator node types available to drag onto the canvas

Components are the execution nodes you drag from the Components panel onto the canvas. Available only in standalone workflows.

Agent​

An Agent node runs a single LLM prompt.

FieldDescription
Agent NameLabel for this step (displayed on the canvas)
Prompt TemplateInstruction sent to the LLM; supports {{parameter_name}} placeholders
LLMDrag an LLM Chat Setting from the LLMs tab (required)
ToolsDrag tools from the Tools tab (optional)
SourcesDrag sources from the Sources tab for RAG retrieval (optional)

An Agent node must have an LLM assigned before the workflow can be saved or run.

Workflow​

A Workflow node delegates execution to a separate standalone workflow.

  • Select the target workflow to invoke.
  • The referenced workflow runs as a sub-step and its output feeds into the next node.

Orchestrator​

Orchestrator node in a standalone workflow — showing Orchestrator Name, Prompt, LLM, and Skills fields

An Orchestrator node coordinates multiple agents or sub-workflows.

  • Routes work to sub-agents based on the task at hand.
  • Useful for complex pipelines where different agents handle different parts of a request.
note

The Components panel (Agent, Workflow, Orchestrator as separate node types) is only available in standalone workflows. In assistant workflows, Orchestrator behaviour is accessed via the Mode selector on the canvas node, not as a separate draggable component.

Output​

The Output node is always present on the canvas.

  • Marks the end of the execution chain.
  • Connect the last component node's output handle to the Output node's input handle.
  • Nodes not connected to the path from Input to Output are not executed.

Sticky note​

Sticky notes are documentation-only elements — they do not execute.

Use them to annotate steps, explain design decisions, or leave notes for collaborators. Sticky notes are preserved when you save and export the workflow.

Attachment tabs​

Attachment tabs appear in the builder panel of both standalone and assistant workflows. You drag items from these tabs directly onto a canvas node to configure it — they do not create new nodes.

TabWhat you dragEffect
SourcesDocument sourceEnables RAG retrieval from that source inside the node
ToolsToolMakes the tool available for the node's LLM to call
SkillsSkillAdds a pre-built skill to the node
LLMsLLM Chat SettingSets the model the node uses (required for Agent nodes)

Assistant workflow panel​

Assistant workflow editor showing Sources, Tools, Skills, LLMs, and Sticky note — no Components panel

The assistant workflow editor has no separate Components panel. Instead, each canvas node has a Mode selector that switches it between Agent and Orchestrator behaviour.

Node Mode: Agent​

Assistant workflow node in Agent mode — Mode dropdown set to Agent, with LLM and prompt fields visible

In Agent mode the node runs a single LLM prompt. The left panel shows the attachment tabs: Sources, Tools, Skills, LLMs, and Sticky note.

Node Mode: Orchestrator​

Mode dropdown on an assistant workflow node showing Agent and Orchestrator options Assistant workflow node switched to Orchestrator mode — left panel adds an Agent block for creating sub-agents

Switch the node to Orchestrator mode to coordinate multiple sub-agents. When Orchestrator mode is active:

  • The node routes queries to specialized sub-agents, each with their own tools and sources
  • The left panel gains an Agent block — drag it onto the canvas to add sub-agent nodes under the orchestrator
  • Each sub-agent can have its own LLM, Sources, Tools, and Skills configured independently

Execution order​

  • Execution follows edges from Input → component nodes → Output.
  • Multiple component nodes can be chained: Input → Agent A → Agent B → Output.
  • Any node not connected to the path from Input to Output is ignored at run time.
  • Component nodes require an LLM; the run is blocked with an error if any LLM is missing.

Tips for clean graphs​

  • Name each component node after its outcome (e.g., "Extract fields", "Summarise", "Draft reply").
  • Keep prompts focused; use the Input node's prompt template for shared boilerplate.
  • Use sticky notes to document non-obvious routing decisions.