How to create an AI agent without code: a step-by-step guide
Last updated:
Try a chatbot on your own website.
Paste your website address, then ask a question about your pages.
The demo reads your homepage and up to four more public pages and answers up to six questions, with no account. Actions, channels and the full knowledge base need a free account.
Free demo · No account or credit card needed
Ready to build? Configure your agent and add it to your site.
What an AI agent is
An AI agent is a language model with three things added: your own knowledge to answer from, instructions that set its role and limits, and tools it can use, such as booking an appointment or looking up a product. A chatbot only talks. An agent can also check something and do something during the conversation.
For a business, that usually means an agent on the website and in messaging apps that answers customer questions from your pages and documents, collects contact details, books meetings and hands a conversation to a person when it should. You do not need to train a model yourself. The builder connects an existing model, such as Claude or GPT, to your content.
Step 1: Give the agent one job
The most common mistake is an agent that is meant to do everything. Start with one job you can describe in a sentence, then add more once the first one works. Look at the questions your team answers by email, phone and chat each week and pick the group that repeats most.
| Job | Knowledge it needs | Actions it needs |
|---|---|---|
| Answer customer service questions | Help pages, shipping and return rules, FAQs | Hand-off to a person, optional ticket in your helpdesk |
| Qualify leads on a service website | Service pages, prices, areas you cover | Contact form in the chat, lead alerts by email |
| Book appointments | Services, durations, preparation | Booking through Cal.com, Calendly or SimplyBook |
| Help shoppers find products | Product pages, size guides, policies | Product search, cart and checkout link, order status |
| Answer staff questions internally | Handbook, processes, Notion or Google Drive | None at first; Slack as the channel |
Step 2: Train it on your own content
The agent can only be as good as the content behind it. Most businesses start with their website, then add files and a few exact answers.
In intoCHAT you click New agent and either pick a template or choose Start from your website. That option reads your homepage and up to four key pages, takes your logo and colours, writes first instructions and trains the agent in a few minutes. Afterwards, add more on the Knowledge tab and click Train agent.
- Website: enter your address, click Find pages and untick pages that do not help customers, such as legal notices or old news.
- Files: PDFs, Word documents and other files with prices, terms or product sheets.
- Text and Q&A pairs: exact answers for questions where the wording matters, such as cancellation terms or delivery times.
- Connected sources: Notion pages, Google Drive files and solved helpdesk tickets, so the agent learns from answers your team already wrote.
Step 3: Write short instructions with clear limits
Instructions tell the agent who it is, how it speaks and what it must not do. Keep them short and specific. Facts belong in the knowledge, not in the instructions, so you can update a price without touching the prompt.
intoCHAT adds its own rules on top of yours: the agent answers from your sources, says so when it does not know, and resists attempts to change its instructions. You can copy the example below and adapt it, or use the free system prompt generator.
Example instructions for a customer service agent
You are the customer service assistant of Example Ltd, an online shop for outdoor gear. - Answer questions about orders, shipping, returns and products, using only the knowledge provided. - Keep answers short: two to four sentences, with a link to the right page when there is one. - Answer in the language the customer writes in. - If the answer is not in your knowledge, say so and offer to pass the question to the team. - Never promise refunds, discounts or delivery dates that are not in your knowledge. - If a customer is upset or asks for a person, collect their email address and hand the conversation to the team.
Step 4: Add only the actions the job needs
Actions turn a chatbot into an agent. Every action is one more thing that can go wrong, so add the ones your job from step 1 needs and nothing else.
- Collect leads: a contact form inside the chat, with alerts to your inbox and an export of the leads.
- Book meetings: connect Cal.com, Calendly or SimplyBook, and the agent offers open times and books them in the chat.
- Hand off to a person: your team takes over the chat in live chat when someone has the inbox open, or the conversation is emailed to them, optionally as a ticket in Zendesk, Freshdesk, HubSpot, Gorgias or Help Scout.
- Search products: for Shopify, WooCommerce, Magento, VTEX, WordPress or a custom shop, the agent searches your live catalogue and shows product cards.
- Call your own API: a custom API action lets the agent check live data, such as stock or an account status, with inputs filled in from the conversation. Client-side actions run a JavaScript function on your page.
- Procedures: step-by-step workflows the agent follows for one task, such as a return request, from the Starter plan.
Step 5: Test it before customers see it
Open the Playground and ask the twenty questions customers ask most often, in the words they use. Check three things: is the answer correct, does it point to the right page, and does the agent admit when your content does not cover something. When an answer is wrong, add a Q&A pair with the right answer and train again.
Then try what a real visitor might do: ask two things at once, write in another language, ask for a discount or try to make the agent ignore its instructions. From the Starter plan you can save these chats as tests with simulated customers. An AI plays the customer, an AI judge grades each chat pass or fail, and you run them again after every change.
Step 6: Put it live on your website and channels
Copy the embed code from the Share tab and paste it before the closing body tag of your site, or use the WordPress plugin. On Shopify you paste the code into your theme. The chat bubble appears with your colours and welcome message.
From the Starter plan, the same agent can also answer on WhatsApp, Facebook Messenger and Instagram, by email and in Slack, with the same knowledge and instructions. It answers the messages people send. It does not send campaigns or message people first.
Step 7: Read the conversations and improve every week
The first weeks show what your content is missing. Read a sample of conversations, look at the questions the agent could not answer and add the missing facts as Q&A pairs or pages. Small weekly fixes improve an agent more than rewriting the instructions.
- Conversations: every chat with its sources, so you can see why the agent answered as it did.
- Questions to improve: questions your content did not cover, grouped by the day they first came up.
- Satisfaction: a rating from 1 to 5 after a live chat, and an optional survey by email when a conversation is solved.
No-code builder or a framework such as n8n or LangChain?
Developers build agents with frameworks and workflow tools such as LangChain, the OpenAI Agents SDK or n8n. That gives full control, but you also build the chat window, the knowledge search, the hand-off and the monitoring yourself. A no-code builder includes those parts and is faster for customer-facing agents.
| No-code builder (e.g. intoCHAT) | Framework or workflow tool (e.g. LangChain, n8n) | |
|---|---|---|
| Time to a first agent | About an hour | Days to weeks |
| Skills needed | None; editing text and settings | Programming or workflow design |
| Chat widget, inbox, hand-off | Included | Built or connected by you |
| Knowledge search over your content | Included, with training from website, files and Q&A | You choose and run a vector database |
| Custom logic | Through API actions and procedures | Unlimited |
| Best for | Customer service, sales and booking on your website and messaging apps | Internal automations and agents deep inside your own systems |
What it costs
You can build and test an agent on intoCHAT’s Free plan: $0, no credit card, with 1 agent, 25 messages a month, 100,000 characters of training content and 1 website crawl a month. That is enough to finish every step in this guide, but not to run a busy website.
Starter costs $19 a month for 3 agents and 2,500 messages, and adds procedures, tests with simulated customers and the WhatsApp, Messenger and Instagram channels. It starts with a 7-day free trial. Pro costs $49 a month for 10 agents and 10,000 messages. Each message a visitor sends counts as one with the default model.
What intoCHAT does not do
intoCHAT is built for agents that talk to your customers in writing. It does not answer phone calls, it has no visual flow builder for rule-based decision trees, and it never messages people first. It is also not a full helpdesk with its own ticketing: hand-offs go to your team by live chat or email, or to the helpdesk you already use. If you need any of these, a different tool will suit you better.