Design n8n-style intelligent workflows directly inside CupCarbon — or from your browser — and connect them to your simulated IoT network. Chatbots, LLM agents, MQTT pipelines and sensor automation, all visual.
Open it from the CupCarbon toolbar (🤖 button). Dark n8n-like canvas: add nodes from the ☰ palette, drag them, connect output ports to input ports, edit parameters in the side panel, zoom and pan. Workflows are saved per project (one JSON file each, kagent-compatible) and you can create as many as you want. Run one or run them all — they execute continuously until stopped, independent of the simulation, with a green running frame, live node flashes and a per-node output inspector.
Every node receives an array of JSON items and applies its function to each item — exactly like n8n. Parameters accept JavaScript expressions evaluated per item in a sandboxed engine:
Hello {{ $json.firstName.toUpperCase() }}
{{ $json.email.split("@")[1] }}
{{ $json.items.length }} items — index {{ $itemIndex }}
The Code node runs full JavaScript with $json, $itemIndex and
$input.all(), in two modes: once per item, or once for all items.
A real LLM in your workflow: choose Anthropic Claude or OpenAI, set your API
key once in the editor's API Keys dialog (stored locally, never hard-coded), and the
agent answers with structured JSON merged into the items —
{"action": "mark", "response": "marking the node"}. Optional window-buffer
conversation memory makes multi-turn chatbots possible, and a configurable rule-based
fallback keeps workflows running without a key or network.
Chat Trigger ─▶ AI Agent (memory: window) ─┬─▶ Chat Response
└─▶ If action == 'mark' ─▶ CupCarbon Send
Workflows and simulated nodes talk both ways:
cup_toagent('temperature 42', 32) in a node's Python script fires every
CupCarbon Read trigger with cupid 32;msg = cup_fromagent(32);Everything also works remotely over MQTT with secure per-session topics (Copy Topic button): the Web Workflow Editor (design visually in the browser, then send to CupCarbon), CupCarbon Studio (console + workflow exchange with visual preview) and the Web Console.