Nodes and their types
A node is a functional building block — one step that performs a specific operation. Flux ships 38 nodes across eight types, and knowing which type you need is most of knowing which node you need.
The eight types
| Type | What it does | Nodes |
|---|---|---|
| Trigger | Starts the workflow — an event, an API call, a schedule, an incoming message | 13 |
| Action | Performs a task mid-flow: calls an API, sends a message | 2 |
| AI | Uses an LLM to process input, generate text, or decide what to do next | 2 |
| Logic | Adds decision-making — branch on a condition | 1 |
| Response | Defines the final output of the workflow or subflow | 10 |
| Tool | A capability an AI node may choose to call, with parameters the model fills in | 7 |
| Attachment | Handles model memory and file or media input — what the AI remembers and reads | 3 |
| Transformation | Modifies, formats or restructures data between nodes | — |
The node palette open, showing nodes grouped by type with a coloured icon for each
Screenshot to be addedType by type
Trigger — how a workflow starts
Nothing runs without one, and the trigger decides what kind of thing the workflow is. A chat trigger makes it a bot; an API trigger makes it an endpoint; an interval trigger makes it a scheduled job. One workflow can carry several.
See Trigger nodes.
AI — the reasoning step
GPT Model or Gemini Model. This is the node that reads context, decides what the request needs, and calls tools if you gave it any. Attach memory so it remembers the conversation and a knowledge base so it can answer from your documents.
See AI nodes.
Action — doing something mid-flow
An action runs every time the flow reaches it. Use one when the step is not optional: always call this API, always send this email.
See Action nodes.
Tool — doing something only if needed
A tool is the same kind of capability, attached to an AI node's tool connector instead. The model reads its description and decides whether this request needs it.
See Tool nodes.
Logic — branching
The Decision node compares a value and sends the flow down a different path. Use it for deterministic branching you do not want a model deciding.
See Logic nodes.
Response — the answer
Where the workflow's output goes: back to the chat UI, out to a messaging channel, or returned to whoever called the API. Match the response node to the trigger — a WhatsApp trigger pairs with a WhatsApp response.
See Response nodes.
Attachment — memory and knowledge
Attachments hang off an AI node and change what it knows: GPT Memory and Gemini Memory carry the conversation forward, and Knowledge Base lets it answer from your documents.
See Attachment nodes.
The shape most workflows take
Almost everything you build is a variation on this:
- A trigger fires.
- Optionally an action fetches something needed every time.
- An AI node reasons over it, with attachments for memory and knowledge and tools it may call.
- Optionally a decision routes the result.
- A response answers.