Chat Trigger
The Chat Trigger turns a workflow into something people can hold a conversation with. Every message they send starts a run, and whatever the last step produces comes back as the reply.
Two ways to use it. Hosted Chat gives you a ready-made chat page on a link you share. Embedded Chat gives you a URL to POST to from your own widget, so the conversation lives inside your product.
Setting it up
Add the trigger
Open Tools in the dock and choose Chat Trigger under Start.
Pick a mode
Hosted Chat is the quickest way to something usable: Glow hosts the page, you share the link. Choose Embedded Chat when you already have a chat interface and only want the workflow behind it.
Decide who may talk to it
Authentication defaults to None, which means anyone holding the link can chat. Glow User Auth admits only members of the team that owns the workflow.
Build the reply
Add the steps that decide what to say. An AI Prompt reading {{ N.chatInput }} is the usual starting point, where N is the Chat Trigger’s step number.
Try it before anyone else does
The step’s Chat tab holds a conversation with your own workflow. Ask it something and watch the reply come back, using a sample message it provides if you would rather not type one.
Go live
The chat URL appears once the workflow is Live. In Draft the panel shows a Go live button instead of a link.
What it passes on
| Reference | What it holds |
|---|---|
{{ N.chatInput }} | The message the visitor just sent |
{{ N.sessionId }} | Identifies one visitor’s conversation across messages |
{{ N.metadata }} | Anything your own widget sent alongside the message |
{{ N.historyText }} | The recent conversation as text, ready to drop in a prompt |
historyText is the one that makes a chat feel like a conversation rather than a series of unrelated questions. Put it in an AI step’s instructions and the model can see what was already said.
Giving it a memory
Conversation memory (messages) sets how many earlier messages of the same sessionId are handed to the workflow as history. Set it to 0 and every message is answered on its own, with no idea what came before.
A longer memory makes for better answers and a longer prompt. Start small and raise it if the model keeps losing the thread.
Keeping the cost predictable
Every message runs the workflow and spends your team’s credits, so an open link is a running cost as well as an open door.
Messages per visitor per minute caps how fast one person can send. A per-visitor rate limit is always applied when Authentication is None, and this setting is how you tighten it. For an internal tool, Glow User Auth is the stronger control: only your team can start a run at all.
Making the page your own
Four settings shape the hosted page:
| Setting | What it does |
|---|---|
| Chat title | The heading at the top of the page |
| Welcome message | Shown before the visitor types. Display-only: it never reaches the workflow |
| Input placeholder | The grey text in the message box |
| Response mode | When Last Step Finishes holds the request open until the workflow completes, then returns the final step’s output |
Running Workflows from Slack
Beyond hosted chat links, team members can interact with workflows directly inside Slack channels:
- Natural Language Triggering: Mention the Glow Slack Bot or Workflow Assistant with instructions (e.g.
@Glow summarize the latest Q3 pipeline leads). - Intent Matching: The assistant matches the request to live published workflows in your workspace.
- Parameter Injection: Key arguments mentioned in the message or thread (e.g., deal names, dates, or assignee emails) are parsed and mapped to the initial trigger fields.
- Execution & Reply: The workflow runs in the background and posts the formatted output back into the Slack thread.
Examples to copy
A support bot over your own documentation
Chat Trigger → AI Prompt. In the instructions:
Answer the question using our product documentation.
Conversation so far:
{{ 1.historyText }}
Question: {{ 1.chatInput }}Set Conversation memory to 10 so follow-up questions like “and what about the Pro plan?” still make sense.
An internal assistant, restricted to your team
Set Authentication to Glow User Auth and share the link internally. Only members of the workflow’s team can open it, and no rate limit is needed for an audience you trust.
A chat window inside your own product
Set Mode to Embedded Chat and POST each message as JSON to the URL from your widget. Put the message itself in a field (a key in the JSON body) called chatInput: it is the only field the endpoint requires. Send the same sessionId with every message from one visitor so their conversation continues; leave it out and each message starts fresh. Add a metadata object for your own fields, such as a customer id or the page they are on, and read them in the workflow as {{ 1.metadata }}.
What’s Next?
- AI Prompt: the step that usually writes the reply.
- Webhook Triggers: for starting a workflow from a system rather than a person.
- Variable Syntax: the full reference for
{{ }}.