chatleadr Docs
Open Chatleadr

What a Workflow Is

By default a bot answers questions from its knowledge base. A workflow lets it act instead: send a document, collect details, call an API, push a lead into ClickUp, or hand the visitor to WhatsApp.

Start FlowStart NodeChatbot Trigger Look up the orderAPI RequestGET /orders/:number Give them the statusChat MessageOrder order_number shipped on… Offer a hand-offChat MessageI couldn't find that order. Shall I… SendAPIFormDecisionFunctionWaitWebhookMenuEnd
A workflow that looks an order up in another system, then answers with what it found.

What a Workflow Needs to Run#

A workflow the assistant can start on its own is three things together.

  1. The nodes. Steps on a canvas: send a message, ask a question, call an API, branch on the answer.
  2. The description. Prose telling the assistant when to run it. This is the only input to that decision. See writing a workflow description.
  3. The parameters. Values the assistant extracts from the conversation and passes to the workflow, such as an email address, a product name or a date.

Let the bot start this is the switch that makes a workflow reachable, and it cannot be turned on without a description. With it off the workflow still exists, and nothing in a conversation reaches it: neither a trigger keyword nor a conversation starter runs a workflow whose switch is off. Every new workflow starts with that switch off.

The assistant chooses The workflow executes Visitor “Can I talk to someone about a quote?” send_brochure no match book_a_call matches: asked to speak to someone check_order_status no match Chosen by reading the descriptions, not by matching words. Form: ask for a preferred day API: create the booking Message: confirm the time Same order, every run.
The assistant reads descriptions and chooses. From the Start node onwards, nothing is decided by a model.

What the Model Decides, and What It Does Not#

The assistant reads every workflow description, picks one or none, and fills the parameters. From the Trigger node onwards, no model decides anything:

  • A Decision node branches on a value, not on a judgement.
  • A Function node runs your code, not a prompt.
  • Anything needing loose interpretation has to happen before the workflow starts, as a parameter the assistant fills in.

How Many Workflows a Bot Can Have#

The limit is set by plan and applies per bot. See plans and limits.

Selection accuracy falls as the list grows, because the assistant compares descriptions and near-duplicates compete. A bot needing more distinct jobs than its plan allows is better split across bots, each with a focused set and its own knowledge base.

Common Questions#

Can the Bot Start Any Workflow on Its Own?#

Only the ones you let it. A workflow carries a switch for that, Let the bot start this, and the dashboard badges those workflows AI when the bot is in AI mode and Keyword when it is in keyword mode. Leave the switch off and the workflow still exists, but nothing in a conversation reaches it: a trigger keyword and a conversation starter both need the switch on.

Can Two Workflows Run in the Same Conversation?#

Yes, one after another. A workflow runs to an End node, the conversation continues, and a later message can select a different workflow. Two workflows do not run at the same time.

What Happens If No Workflow Matches?#

The bot answers normally from its knowledge base. Selecting nothing is a valid outcome, and the common one: most messages are questions rather than requests for an action.

Do Workflows Work on WhatsApp and Instagram?#

Yes. The same workflows run on every channel the bot is connected to, and what changes is how each node renders. Forms are the largest difference: they render inside the chat on the web and arrive as a secure link on WhatsApp and Instagram. See what differs per channel.