Node reference

A workflow is a graph of nodes connected by edges. Everything MAIA Plus can do inside a workflow is a node type: start a run, read a ticket, call a model, branch, loop, write a row. There are 26 registered types, and this page is the map.

How to read the node pages

Every node page follows the same structure: purpose, screenshot of the panel, configuration fields, input and output handles, expression support, errors and effects, and a worked configuration example.

Element	What it means
Configuration table	Field name, type, default and effect, taken from the schema this version ships
Example configuration	A schema-valid configuration object. The field names and types are correct for this version; the identifiers are placeholders and must be replaced with your own
Expected result	The behaviour the node produces when it runs with that configuration — not a recorded run from a particular installation

Two conventions are worth knowing before you rely on an example:

• Examples are not recordings. They are validated against the schema, so they will not be rejected for an unknown field, but the values must match your installation — a table id, a tool key or a model profile that does not exist in your environment fails at run time.
• Expected results describe behaviour, not data. The shape of a result is stable; its content depends on the ticket, the knowledge base and the model that produced it.

What a node is

Part	Meaning
Type	The registered kind (trigger_webhook, action_sdp, logic_if, ai_agent, …). Determines behaviour and configuration.
Label	The name you see on the canvas. Free text; does not affect execution.
Configuration	Type-specific fields, edited in the node panel.
Inputs and outputs	Named handles. Some nodes have one output, branch nodes have several.
Effect class	Read, write, or control flow. A write node changes another system.

The logic_loop_boundary type is internal: schema-v3 loops need it, but it is not something you add from the palette.

Read a node card

Every card shows the same three things:

1. Title — the label you gave it.
2. Type badge — WEBHOOK, ticket_read, SINGLE COMPLETION, KNOWLEDGE VAULT (RAG), AUTONOMOUS LOOP, TRUE / FALSE, FINAL OUTPUT. The badge is the contract; the title is decoration.
3. One-line purpose — the description written by the author of that node.

Selecting a node opens its toolbar: Run to this node, Add Breakpoint, Configure parameters, Duplicate this node, Delete node.

Palette groups

Group	Node types
Triggers	Manual Trigger, Webhook Trigger, Schedule Trigger
ServiceDesk	SDP Tools, Knowledge Vault (RAG), SDP Action, DataTables, SDP Database Query, Incident Search
Control flow and logic	IF Condition, Switch, Loop Over Items, Try / Catch, Merge, Subworkflow Call, Return / End
Data and JSON transformation	Data Transform (Set)
AI and LLM	LLM Call, AI Agent, Prediction Models, Predictions, Knowledge (RAG)
Advanced integrations	JavaScript Code, MCP Tools, HTTP Request

Provider nodes versus action nodes

The most common source of confusion, and the reason two nodes can look identical in the palette:

• Provider nodes hand capability to an AI Agent without executing anything: SDP Tools, Knowledge Vault (RAG), Prediction Models, MCP Tools. They connect to an agent’s tools inlet, not to the main flow.
• Action nodes execute immediately: SDP Action, DataTables, SDP Database Query, Incident Search, Knowledge (RAG), Predictions. They sit on the main flow and take an edge.

An agent with no provider node has no tools. An action node with no incoming edge never runs.

Node pages

Each type has its own page with the configuration panel, parameters and failure behaviour.

By category

• Triggers — Manual, Webhook, Schedule
• ServiceDesk — SDP Tools, Knowledge Vault (RAG), SDP Action, DataTables, SDP Database Query, Incident Search
• Logic — IF Condition, Switch, Loop Over Items, Try / Catch, Merge, Subworkflow Call, Return / End
• Data — Data Transform (Set)
• AI — LLM Call, AI Agent, Prediction Models, Predictions, Knowledge (RAG)
• Integrations — JavaScript Code, MCP Tools, HTTP Request

Limits and failure modes

• Wrong handle connected → a branch node activates an output with no downstream connection; the run ends early and reports success.
• Provider node on the main flow → it never feeds the agent and the agent has no tools.
• Renamed label → nothing breaks; do not debug by label.
• Unknown type in an imported graph → the node renders as unknown and the graph cannot be activated.
• Loop without a bound → use maxIterations; an unbounded loop is a runtime hazard, not a feature.

Related topics

Workflow designer · Expressions and context · AI agent nodes · Testing and debugging