What the Chatbot node does
The Chatbot node adds an AI chat widget to your WordPress site. You configure it visually on the canvas, then deploy it anywhere with a shortcode.
Every Chatbot node is an AI agent. It can hold a normal conversation, and it can also take actions: run your workflows, search your posts and WooCommerce products, look things up in a knowledge base, and hand the conversation to a real person. You do not turn “agent mode” on or off. If you give the chatbot capabilities, it uses them. If you give it none, it behaves exactly like a simple reply-only chatbot.
The Chatbot node runs on your WordPress site (local only). Its AI replies are billed either to your connected account credits or to your own provider keys, which you choose per node on the Model tab. See BYOK keys vs keyless credits.
Add a Chatbot to a workflow
- Open a workflow and add the Chatbot node from the node palette (Communication group).
- Pick a model and set your instructions.
- Optionally add actions, a knowledge base, human handoff, and design.
- Save the workflow. A shortcode appears on the node.
- Paste the shortcode into any post, page, or widget area. See Deploy the chat widget.
The configuration tabs
The node splits its settings across tabs.
Model
- AI source. Choose how this chatbot’s AI is powered: Credits meters against your connected account with no API keys, or My API keys uses the provider keys from Settings (free and unlimited). This choice overrides your account default for this chatbot only.
- Model. Pick from OpenAI (direct), Anthropic, or hundreds of models through OpenRouter.
- OpenAI tools. When you use a direct OpenAI model, you can enable provider-native web search and file search.
- Knowledge base. Attach a knowledge base so the chatbot retrieves relevant excerpts each turn and answers from your own content. See Knowledge Base (RAG).
- Memory. Decide who the chatbot remembers and how much history it keeps verbatim. See File uploads and memory.
- Model parameters. Temperature, top-p, and maximum response length.
Instructions
This is the single system prompt for the chatbot: its role, personality, goal, and rules. It is the one place you tell the agent who it is and how to behave.
You can also connect other nodes into the Chatbot node. Their output is fed in as knowledge the chatbot can use.
Under Advanced, set the maximum number of model turns per message when the agent uses tools. This cap is ignored for simple single-reply chats.
Actions
Actions are things the agent can do, plus the full capability list. Each action you add creates a connectable output handle, so a chat can trigger the rest of your workflow. This tab also holds the Capabilities table, where you enable each tool, require approval before it runs, and set a rate limit. See Agent actions and tools.
Handoff
Turn on human handoff and pick a provider: the built-in Operator Inbox (Native), Chatwoot, Zendesk Sunshine Conversations, or Intercom. See Human handoff.
Behavior
- Initial message. A fixed greeting, or an AI-generated one based on the page.
- History and rate limiting. Cap stored history and limit how many messages a visitor can send in a time window, to protect against abuse.
- File uploads. Off by default. Let visitors and agents share documents and images.
- Advanced. Typing indicator, page context, sound effects, auto-open delay, persist history across reloads, source citations, and the collapsible agent activity line.
Design
Control theme (light, dark, or custom), position, bot identity, typography, dimensions, colors, quick reply buttons, the send button, and the footer. See Deploy the chat widget.
How the agent runs
When the chatbot has a real capability configured (one or more actions, or agent-decided handoff), it runs a bounded tool loop: the model can call a governed tool, read the result, and continue until it has an answer or a guard stops it. Guards include the maximum steps, a per-run tool cap, and loop detection. With no capabilities configured, the node makes exactly one model call, identical to a classic chatbot reply.
Passing collected data downstream
Each action is an output handle whose id is the action id. Collected fields are referenced downstream as [[fieldName] from {nodeId}:{actionId}]. Every edge from the Chatbot node to a downstream node must connect from the action handle. See Data flow and variables.
Example
A support bot named Ava:
- Model: gpt-4o-mini, AI source Credits.
- Instructions: “You are Ava, a friendly support agent for Acme. Resolve billing questions and hand off to a human when a refund is requested.”
- Action: “Newsletter Subscription” collecting a required email field, wired to a Send Email node.
- Handoff: Native, agent decides plus keyword.
Writing tip for links
When you want the chatbot to send a clickable link, tell it in the system prompt to write links in Markdown, like [Read the guide](https://example.com/guide). The widget renders Markdown, so a plain URL is not guaranteed to be clickable.