WhatsApp Flow Agent

A visual, drag-and-drop conversation builder for WhatsApp agents.

WhatsApp Flow Agent

The WhatsApp Flow Agent is a visual, drag-and-drop conversation builder designed for WhatsApp agents. It allows you to create deterministic, no-code, and fully auditable conversational flows, ensuring predictable interactions for structured use cases.

How It Works

Instead of an AI freely generating replies, you construct a graph using 19 different node types, including messages, buttons, forms, conditions, API calls, and AI/human handoff. The bot then guides the user through this predefined graph step by step. Every inbound message on an agent's number configured as a Flow Agent will be driven by this graph, providing a structured and auditable conversational experience.

How to Use

1

Step 1

Open the WhatsApp Agent settings, create/select an agent, and set its type to "Flow Agent".

WhatsApp Agent settings with type set to Flow Agent
2

Step 2

Drag a node from the sidebar onto the canvas and connect it to the Start node (or another node) with an edge.

Dragging a node onto the Flow Builder canvas and connecting it to the Start node
3

Step 3

Click a node to open its config panel and fill in its content.

Node config panel open for editing
4

Step 4

Add the nodes your conversation needs. The available node types are:

  • Text Message — sends one or more plain-text messages, then auto-advances.
  • Media — sends an image/video/document with a caption, then auto-advances.
  • Template — sends an approved WhatsApp template, then auto-advances.
  • Reply Buttons — sends up to 3 quick-reply buttons and waits for a tap.
  • List / Menu — sends a WhatsApp list message and waits for a selection.
  • Multi-select — asks the user to pick multiple numbered options (e.g. "1,3").
  • Text Input — free-text prompt, stored into a variable; can branch on a regex match.
  • Date / Time — prompts for and parses a date/time (fuzzy parsing, re-prompts if invalid).
  • Location — asks the user to share a WhatsApp location pin.
  • Image Input — asks for a photo and stores it.
  • CTA / URL — sends a message with a link button, then auto-advances.
  • Form — walks a list of fields one at a time, each with its own validation (text/date/number/email/phone), optional "skip" for non-required fields.
  • Delay / Nudge — waits N minutes; branches depending on whether the user replied before the deadline.
  • Condition — branches true/false on a variable (equals/contains/greater-than/date comparisons/"TODAY" magic keywords).
  • API / Webhook — calls an external API, maps the response into variables, branches on success/error.
  • Jump to AI — hands the conversation to an AI agent (see step 6).
  • Human Handoff — hands the conversation to a human (see step 5).
  • Tag Contact — applies tags to the matching contact.
  • End Chat — sends a farewell and closes the conversation as resolved.
Flow Agent node types sidebar
5

Step 5

Add a Human Handoff node to pause the bot and route the conversation to a person.

  • Optionally assign a specific staff member — they get an email with a link to the inbox.
  • The bot stays paused until a staff member marks the conversation done (also closes it), or the flow hits another Human Handoff node.
  • An AI agent can trigger the same handoff mid-conversation via its own "assign to human" action (to a person, or round-robin) — it behaves identically either way.
6

Step 6

Use Jump to AI for a two-way handoff between flow and AI.

  • A "Jump to AI" node sends an optional message and hands the conversation to a chosen AI agent — every reply after that comes from the AI, until it hands back or ends the chat.
  • The reverse also works: an AI agent can jump into the flow at any node (not just Start), even continuing in the same turn.
  • This lets a scripted flow handle structured steps while AI handles the open-ended parts.
7

Step 7

Add an End Chat node to close the conversation.

  • Sends a farewell message and marks the conversation resolved.
  • Collected variables (name, email, form answers, etc.) carry over if the same contact messages again later.
  • Closing a conversation manually from the inbox instead marks it abandoned, not resolved — so the two stay distinguishable in analytics.
8

Step 8

Buttons and template Quick-Replies always take priority, from anywhere in the graph.

  • Tapping a button or list item is honored immediately, no matter what step the flow is currently on — it's matched to its own node, not just the node the flow is waiting on.
  • The same applies to a Quick-Reply on a WhatsApp template — it can jump straight to a configured point in the flow.
  • Result: users can re-engage from an old message at any time without getting stuck.

Benefits

  • Deterministic, predictable conversations: Ideal for structured use cases like bookings, lead qualification, and support triage where free-form AI replies are undesirable.
  • No-code: Support and operations teams can build and edit flows visually without requiring engineering involvement.
  • Full native WhatsApp interactivity: Leverage quick-reply buttons, list menus, forms with per-field validation, date/time and location capture, image capture, multi-select, and delayed nudges with timeouts.
  • Can call external APIs mid-flow and branch on the response: Integrate with external systems to fetch or send data during a conversation.
  • Two-way handoff with AI agents and with human staff: A flow can escalate to AI reasoning or a human agent and seamlessly return to the flow.
  • Immediate response to buttons/templates: A tap on a button or a template Quick-Reply is honored immediately, regardless of the flow's current state, preventing users from getting stuck mid-form.

AI Chat Agent, WhatsApp Campaigns, WhatsApp Templates, Contacts, WhatsApp Inbox

Screenshots

WhatsApp Flow Agent screenshot 1

Related Articles

IntroductionAI-Getting StartedWhatsapp SettingsAPI