Table of Contents

How to build an agent in Landbot

Desirée M Updated by Desirée M

You can build a Landbot agent in four ways: describe it and let Build it for me create the flow, start from a template, build it block by block from scratch, or ask an AI assistant such as Claude to build it with the Landbot MCP or the Landbot Plugin. Whichever way you choose, the agent opens in the same visual builder, where you can change anything before you publish it.

Landbot home page with the What should your agent do? box, template cards and the Create agent button

Which way should you build your agent?

Each way gets you to a working agent; the difference is how much of the flow you design yourself.

● Build with AI tools: describe the agent to your AI assistant or AI coding tool, which builds it in your Landbot account through the Landbot MCP or the Landbot Plugin.

● Build it for me: describe the agent in your own words and Landbot's AI writes the whole flow, ready for you to adjust.

● Use a template: start from a proven flow, such as Lead Generation, Customer Support or FAQs, and adapt it to your business.

● Start from scratch: begin with a blank canvas and decide every block, message and condition yourself.

If you'd rather not build it yourself, select Let a Landbot specialist build it after Create agent: Landbot's team designs, configures and maintains the agent for you.

How to build an agent with AI tools

To build an agent with an AI tool, connect it to your Landbot account with the Landbot MCP or the Landbot Plugin, then describe the conversation in everyday words. The AI tool creates the agent in your account, and you don't need to know block names or how the builder works.

● The Landbot MCP is a connection you add to an AI assistant or AI coding tool, such as Claude, Claude Code, Codex or Cursor. You sign in to Landbot once, and nothing is installed on your computer.

● The Landbot Plugin is an add-on for Claude Code and the Claude desktop app. It uses the same connection and adds a guided build: it styles the web chat if you describe a look, and in the Claude desktop app it tests every path of the conversation for you.

The result is a normal Landbot agent: you can open it in the builder, change any block by hand and read its conversations in the inbox. The AI tool asks for your approval before publishing an agent and before changing one you already have.

To compare the two and pick one, see What are the Landbot MCP and the Landbot Plugin?. Once connected, How to build, test and publish an agent with the Landbot MCP or the Landbot Plugin walks through your first agent.

How to create a new agent

To create a new agent, select Create agent on the home page (or Build an agent in Agent builder), then choose the channel and where to start.

1️⃣ Select Create agent. In Let's start building!, choose Start from scratch.

Let's start building dialog with the options Start from scratch and Let a Landbot specialist build it

2️⃣ Pick the channel your visitors will use: Web, WhatsApp, Messenger or API Agent.

Channel choice in Landbot with Web, WhatsApp, Messenger and API Agent

3️⃣ Choose where to start: Build it for me!, Start from scratch or Use a template.

Landbot dialog with three ways to start: Build it for me, Start from scratch and Use a template

The builder opens with your new agent. With Start from scratch, the canvas holds only the Starting point, ready for your first block.

Landbot builder with a new agent showing only the Starting point block

How to build an agent from scratch

To build an agent from scratch, add blocks after the Starting point and connect them in the order the conversation should follow. Each block does one job, such as sending a message, asking a question or checking an answer, and it runs only when a visitor reaches it.

In this example, a yoga studio agent, Lotus Yoga Studio, greets visitors with a Buttons block that offers two options:

● Book a free class, the visitor goes on to an Ask for a name block: "Great choice! What's your name?"

● See class times, a Messages block answers: "Classes run Monday to Saturday, from 7 am to 8 pm. Come as you are, we have mats for everyone! 🧘"

Landbot canvas with Starting point, a Buttons block, an Ask for a name block and a Messages block

The rest of the Lotus Yoga Studio flow asks which style the visitor wants to try, their email and how many classes a week they'd like, then recommends a pass. Each of those steps is one block, connected to the next.

Complete Landbot flow for a yoga studio agent, from the Starting point to two closing messages
On Web, the first block after the Starting point must ask the visitor something, such as a Buttons or Ask for a name block. On WhatsApp, any block can start the flow.

How to add and connect blocks

To add a block, click or drag the green dot on the output of a block, then pick the new block from the menu. Search for a block by name, or browse the categories: AI, Messages, Questions, Logic, Integrations, Inbox & Builder tools and Low code. The new block arrives already connected to the output you started from.

Landbot add-block menu with search, Build it for me and block categories, AI category open

To connect two blocks you already have, drag the green dot of an output onto the other block. To remove a connection, hover over it and select the bin icon. The + icon under the bin adds a new block in the middle of that connection.

Landbot connection on hover with a bin icon to delete it and a plus icon to insert a block
Keep blocks apart rather than stacked on top of each other: a tidy canvas is easier to follow when you come back to the flow. The Builder Interface Tour explains the canvas tools for zooming, undoing and moving around.

What kinds of blocks can you add?

Landbot blocks fall into a few families, and a typical agent uses several of them: messages to talk, questions to listen, logic to decide and integrations to pass the answers on.

Message blocks

Message blocks send text, images, videos and other media to the visitor without waiting for an answer. Use them to greet, explain or confirm, such as the class times in the Lotus Yoga Studio agent. See the Messages library for every message block.

Question blocks

Question blocks ask the visitor something and wait for the answer: through buttons, free text, a name, an email, a number, a date and more. Many of them check that the answer fits, so Ask for an email only accepts an email, and each one saves the answer in a Field. See the Question blocks library for every question block.

Logic blocks

Logic blocks work behind the scenes: they check answers, branch the flow, jump to another part of it, count or calculate, without the visitor seeing them. See the Logic & Technical blocks library for every logic block.

Integration blocks

Integration blocks connect your agent to other apps, such as sending an email, adding a row to Google Sheets or booking a meeting in Calendly. Find them under Integrations in the add-block menu, and see the Integrations library for how to set up each one.

Landbot add-block menu with the Integrations category open and its integration blocks
AI Agent block

The AI Agent block lets an AI hold an open conversation inside your flow: it answers questions from its Knowledge Base, follows your instructions, saves what the visitor shares in Fields, and hands the conversation back to the rest of the flow through the outputs you define. Find it under AI in the add-block menu, and see the AI Agent block for full setup instructions.

How do Fields keep the visitor's answers?

Fields keep the visitor's answers so the agent can use them later in the conversation, such as the visitor's name in a closing message. Each question block saves its answer in the Field set in Save user answer in the field.

Many question blocks come with a Field already, so an Ask for a name block saves the answer in name, and you can create your own Fields for anything else you need to remember.

In the Lotus Yoga Studio agent, the Ask for a name block saves the answer in name, and the last message uses it: "Our Unlimited pass is the best fit for you, Omar Rossi!"

Ask for a name block panel saving the answer in the name Field, next to the flow on the canvas

To learn how to create, edit and reuse Fields, see Fields.

How do blocks connect to the next step?

Blocks connect to the next step through their outputs, the dots on the right of each block. Most blocks have a single output, while blocks that branch have one output per path: the green and red outputs of a Conditions block, or one output per button on a Buttons block.

What do the green and red outputs do?

The green and red outputs split the flow in two: the green output is the path visitors follow when a condition is true, and the red output is the path they follow when it's false. The Conditions block and other logic blocks that check something have them.

In this example, the Lotus Yoga Studio agent asks how many classes a week the visitor would like, and a Conditions block checks whether @classes is GREATER THAN 2:

● Green output (true), more than 2 classes a week: "Our Unlimited pass is the best fit for you, @name! We'll email the details to @email. See you on the mat! 🙌"

● Red output (false), 1 or 2 classes a week: "Our 10-class pack is a great fit for you, @name! We'll email the details to @email. See you on the mat! 🙌"

Conditions block checking if classes is greater than 2, with green and red outputs to two messages
What does the Any of the above output do?

On Web, the Any of the above output of a Buttons block is the path every button follows when that button has no connection of its own. It saves you from connecting each button separately when they all lead to the same next step.

In this example, the Lotus Yoga Studio agent asks "Which style would you like to try first?" with the buttons Hatha, Vinyasa and Yin:

● Any of the above, whichever style the visitor picks: the agent saves it in style and asks "What's your email? We'll send your class details there."

Buttons block with three style buttons where only the Any of the above output connects to the next block

For more uses, such as multiple-choice questions, see Any of the above output, and the Buttons block for every button setting.

How to test and publish your agent

To check your agent before visitors see it, walk every path of the conversation: answer each question, pick each button and make sure each ending makes sense. In the Lotus Yoga Studio agent, a visitor who asks for 4 classes a week gets the Unlimited pass message with their own name and email. How to test your agent covers every way to try it.

Landbot web chat recommending the Unlimited pass to Omar Rossi after he asks for 4 classes a week

To put your agent live, select Publish in the top bar of the builder. Visitors get the published version, and Landbot gives you a share link to the live agent.

Landbot dialog confirming the agent is live, with a share link and a Customize Look and Feel button

What to do after you publish your agent

Once your agent is live, put it where your visitors are, give it your brand's look and see how the conversations go:

● Share it or add it to your website: How to launch your agent and How to embed your agent into your website.

● Make it look like your brand: select Customize Look & Feel in the publish dialog, or ask your AI tool to do it with the Landbot Plugin or the Landbot MCP.

● See how it performs: the Analyze section shows how visitors move through your flow.

You're all set! 🎉 Your first agent is live.

Was this article helpful?

Builder Interface Tour

Managing Data in Your Agent: A Guide to Using Fields

Contact