Chat Trigger

Chat Trigger

Run the flow on each chat message from the user.

What it does

Turns a workflow into a chatbot. Instead of running once, the Chat Trigger opens a chat window, and every message the user sends runs the workflow once, with that message as the trigger’s output (chatInput).

The trigger itself never writes a reply. The reply comes from a downstream agent that uses the same Chat Memory as the trigger: the agent saves the conversation to that memory, and the chat window shows whatever the memory contains.

When to use it

• Build an assistant you can talk to from the canvas while designing a workflow.
• Put a chat bubble on specific websites (for example a support helper on your internal tools), using URL Patterns.
• Give an agent a running conversation, so each message can build on the previous ones.

Required setup

A Chat Trigger needs three things wired together:

1. A Chat Memory connected to the trigger — its memory input is required, and the run fails with a clear error if it’s missing. Use Local Memory (pick a conversation) or Persistent Chat Memory (set a conversation key).
2. A downstream agent that uses the same memory node — Tools Agent or RAG Agent. Connect the memory to the agent’s memory input as well as to the trigger’s.
3. The agent’s prompt set to the chat message — for example {{ $json.chatInput }}.

Window Buffer Memory doesn't work here

Window Buffer Memory only lives for the length of one run and isn’t saved anywhere, so the chat window has nothing to display. The chat shows “no Chat Memory connected” in that case.

Inputs and settings

Setting	Notes
Assistant name	Title shown in the chat header. Default: Assistant. On web pages, falls back to the workflow name when empty.
Icon	Emoji or image URL shown in the chat header. Optional.
Greeting	Message shown before the conversation starts. Default: How can I help you today?
Input placeholder	Placeholder text of the message box. Default: Type a message…
Starter prompts	Suggested prompts shown before the conversation begins.
Allow clearing the conversation	Shows a button that deletes the conversation’s messages. Default: on.
URL Patterns	Pages where the chat bubble appears, in the same format as Page Load. Default: <all_urls> (every page).

Outputs

Each message produces one run whose trigger output is:

Field	Description
triggerType	Always chat.
chatInput	The message the user typed.
sessionId	The conversation id of the connected Chat Memory, as a string.
triggeredAt	ISO timestamp of the message.
url, title, tabId	The page the chat was opened on. Only set when chatting from a web page, not from the canvas.

How a message turns into a reply

1. You send a message. The workflow runs, starting from this Chat Trigger only, with your text as chatInput.
2. The agent saves your message. Before calling the model, the agent writes your message to the Chat Memory as a user message.
3. The agent answers. It reads the earlier conversation from the memory, calls the model (and any tools), and gets a reply.
4. The agent saves its reply. The reply is written to the same memory as an assistant message. If an output parser is attached, the structured result is saved as JSON text.
5. The chat window shows it. The chat window doesn’t receive the reply from the run. It re-reads the conversation from the memory:
   • On the canvas, the window updates as soon as each message is saved, so your message appears first and the reply follows.
   • On a web page, the chat checks the conversation every 1.5 seconds, so messages can appear with a short delay. The “thinking” indicator stops when a new assistant message arrives.

Because the conversation lives in the memory, it’s also visible on the chat memories page, and any other workflow using the same memory sees the same messages.

Where you can chat

On the canvas

Running the workflow opens a chat popup instead of running once. This happens when:

• the Chat Trigger is the workflow’s only trigger and you press Run,
• you pick the Chat Trigger in the trigger picker, or
• you press the node’s run button (Test Step).

On web pages

When the workflow is activated, a chat bubble appears on every page matching one of its URL Patterns. If several chat workflows match the page, the bubble lets you pick one. You can also right-click and choose Open chat to open it on any page.

The run still happens in the AWFlow app tab: the extension opens that tab if needed, runs the workflow there, and passes the page’s url, title and tabId to the trigger.

Example workflow

1. Add a Chat Trigger.
2. Add a Local Memory, pick (or create) a conversation, and connect it to the trigger’s memory input.
3. Add a Tools Agent after the trigger, connect a chat model, and connect the same Local Memory to its memory input.
4. Set the agent’s prompt to {{ $json.chatInput }}.
5. Press Run and start chatting.

Troubleshooting

• “No Chat Memory connected” / chat disabled — connect a Local Memory with a conversation selected, or a Persistent Chat Memory with a conversation key, to the trigger’s memory input.
• Nothing appears, not even my own message — the trigger doesn’t save messages; the agent does. Check that an agent runs on every message and that it uses the same memory node as the trigger.
• The agent answers but the chat stays empty — the agent is using a different memory (or a different conversation / key) than the trigger.
• The agent answers something unrelated — its prompt isn’t set to the chat message. Use {{ $json.chatInput }}.
• Each message appears twice, or several replies show up — every agent connected to the same memory saves its own message and reply. Use one agent per conversation.
• On a web page, the “thinking” indicator never stops — the run failed before a reply was saved. Press stop, then check the run in the app’s execution history.
• Stop doesn’t cancel the run — stop only resets the chat window; the run keeps going and its reply appears on the next refresh.
• No chat bubble on a page — check that the workflow is activated and that one of the URL Patterns matches the page.

Related nodes

• Local Memory
• Persistent Chat Memory
• Tools Agent
• RAG Agent
• Memory & Context