Getting started

This guide takes you from a brand new chatbot to one that answers your customers. You don't need any technical knowledge. Steps 1, 2, 3 and 8 are enough for a working chatbot; come back to the others whenever you are ready.

1. Create a chatbot
2. Add your knowledge
3. Shape how it answers
4. Choose the model
5. Connect your tools
6. Draw a workflow, if you need one
7. Try it
8. Put it on your site
9. Add chat apps
10. Watch it and improve it

1. Create a chatbot

On your dashboard, press New chatbot and give it a name, such as your company's name or “Support”. The name is just for you, so you can tell your chatbots apart. Your customers won't see it unless the chatbot mentions it.

Your new chatbot starts as a draft. While it is a draft, only you can talk to it, inside the dashboard, so you can set it up and test it in peace. Nobody else sees it until you put it live in step 8.

Down the left side you'll see the chatbot's sections, from Knowledge to Evaluate. This guide goes through them in the same order.

2. Add your knowledge

Every chatbot needs to know about the business it speaks for. The AI behind it already knows a lot about the world in general, but nothing about you: what you sell, what it costs, when you are open, how returns work. Your knowledge base is where you teach it. You add your own documents and web pages, and the chatbot answers your customers from them, not from guesses.

A good way to start is to think of the questions your customers ask most, and add whatever already answers them. For example:

• An online shop: product pages, the delivery and returns policy, size guides.
• A software company: help articles, the pricing page, setup guides.
• A clinic or a salon: your services and prices, opening hours, how to book.

Under Knowledge there are four ways to add it:

• Files: upload PDF, plain text, Markdown or HTML files. On the Free plan you can add up to 10 MB in total.
• Website: enter your website's address and press Fetch pages. ChatFrom finds your pages, you tick the ones you want, and they are added. When your site changes, add the page again to update it. Pages that haven't changed cost nothing.
• Text: paste in anything that isn't written down anywhere else, like your opening hours or a short policy.
• Q&A: write a question and the exact answer you want the chatbot to give. This is handy for the questions you get every day, and for anything your documents explain badly.

Every answer shows which of your sources it came from. Your customers can check it, and if an answer is ever wrong, you can see where it came from and fix it there.

How sources are split

Your chatbot doesn't read a whole document for every question. ChatFrom cuts each source into small pieces, and when a question comes in, the chatbot reads only the few pieces that fit it best. The standard way of cutting works well for help pages and short articles, so most people never need to change it.

If your documents are long or unusual, such as contracts, a book or product manuals, open Change on How sources are split, then Need help?. Describe your documents in a sentence or two and press Set it up for me, and ChatFrom picks the settings for you. Sources you added before keep their old pieces until you cut them again, which the card offers to do with one click.

3. Shape how it answers

Next, tell your chatbot how to behave: who it works for, who it talks to, what tone to use, and what it should and shouldn't do. You write this under Behaviour, as instructions the chatbot reads before every answer. You may hear these called the system prompt.

You don't have to write them from scratch. Open Need help? at the top, describe your business and what you want the chatbot to do, and press Write my setup. ChatFrom writes the instructions and fills in the settings below them. Nothing is saved until you have read it all and pressed save, so you can change anything first. This uses a few credits from your monthly allowance, and it tells you how many before you start.

If you prefer, you can write the instructions yourself or start from a template. Either way, it's worth going through the settings below them:

• Personality: how formal or friendly it sounds, and how much detail it gives.
• Language: it answers in the visitor's own language, unless you choose one.
• When it does not know: what it says when your knowledge has no answer, and whether it then passes the conversation to your team.
• Guardrails: topics it should politely decline.
• Example answers: a few questions with the answers you would give, so it learns your style.
• Welcome and suggested questions: the first message visitors see when they open the chat, and a few questions they can tap.

4. Choose the model

The model is the AI that writes your chatbot's answers. Under Model you can choose from many models, each with its price. Larger models handle hard questions better but use more of your monthly credits for each answer. Smaller ones are quicker and cheaper. The page shows roughly how many answers a month your plan gives you with the model you pick, so you can compare.

If you're not sure, keep the model your chatbot starts with for now; you can change it at any time. You'll also find a few settings for how varied and how long answers can be, and under Advanced, how the chatbot searches your knowledge. You can leave those as they are.

5. Connect your tools

Knowledge lets your chatbot answer questions. Tools let it do things for your customers, such as checking an order, booking a meeting or passing the conversation to your team. The chatbot decides by itself when a conversation needs a tool, and every time it uses one, you can see it in the conversation log.

Built in

Under Tools, three are ready to switch on:

• Hand off to a person: when the chatbot can't help, it passes the conversation to your team by email, Slack or a ticket.
• Callback requests: a customer can ask for a call back, at a time when you take calls.
• Memory: it remembers what a customer told it before, so they don't have to repeat themselves next time.

Connections

You can also connect the services you already use. Each one has its own guide that walks you through it:

• Shopify: customers can check where their order is, with the tracking link, and ask for a return, which you then approve in your store.
• Stripe: customers can see, cancel or restart their subscription, once they have confirmed their email address.
• Google Sheets: every lead, booking or request is added as a new row in a sheet of your own.
• Cal.com and Calendly: the chatbot offers your free times and books the meeting.
• Your API: for developers. The chatbot can call your own system, and the keys it needs stay safely on our server.
• Your tool server: for developers. If you run an MCP server, the chatbot can use its tools on behalf of your signed-in users.

6. Draw a workflow, if you need one

Most chatbots never need this step. Out of the box, your chatbot handles every message the same way: it looks in your knowledge, writes an answer, and uses one of your tools whenever it needs to. This is the Default loop, and it works well for most businesses.

A Workflow is for when certain conversations should go the same way every time. For example, you may want the chatbot to always ask for the order number before it looks up an order, or to pass every refund request straight to a person. You draw these steps yourself, like a flowchart.

1. Under Workflow, choose Workflow. You start from the default loop's steps, so you build on what already works.
2. Add steps, add a ready-made path such as Book a meeting, or start from a template. A Condition step works out what the visitor wants and sends each kind of request its own way. Ask the Visitor asks a question, with buttons to tap or a box to fill in. Tool Call uses one of your tools, and Human Handoff passes the conversation to a person.
3. Save. Nothing changes for your customers until you do. If you replace a workflow later, the old one is kept under Earlier versions, so you can bring it back.

7. Try it

Before your customers meet your chatbot, have a chat with it yourself. Press Try it at the top of any chatbot page. It opens a chat with your chatbot exactly as it is right now, even while it is still a draft.

Ask the questions your customers usually ask, and look at each answer and the sources under it. If something is off, go back to Knowledge or Behaviour, change it, and try again.

8. Put it on your site

When you are happy with it, it's time to put it in front of your customers. Under Deploy there are four steps:

1. Allowed domains: enter your website's address, such as example.com. The chatbot only works on the sites you list, so nobody else can put it on theirs.
2. Appearance: choose its colour and how the message box looks, whether it remembers returning visitors, and whether to show “Powered by ChatFrom”.
3. Website widget: copy the short piece of code and paste it into your website, just before </body>. Most website builders have a setting for adding code like this. If someone else looks after your site, send it to them; it only takes a minute.
4. Go live: from now on, your chatbot answers visitors on your website, on its share link and in any chat apps you have connected. Press Take offline whenever you want to turn it back into a draft.

You also get a Share link: a page of its own with your chatbot on it. Send it to someone or put it in an email, which is handy if you don't have a website or want to share the chatbot quickly.

9. Add chat apps

Your customers don't only reach you through your website. Under Channels you can put the same chatbot in the apps where they already write to you. It answers with the same knowledge and tools, and every conversation ends up in the same place. The Channels guide explains how this works, and each app has its own step-by-step guide:

• WhatsApp: on your business's own number.
• Telegram: a bot people can message, on their own or in a group.
• Slack: an app in your workspace, for direct messages and threads.
• Discord: a bot in your community's server, and in direct messages.
• Email: forward your support address to it, and it replies in the same email thread.
• Crisp: it answers in the Crisp chat you already have, and hands over to your team when needed.

Sometimes the chatbot needs to know who it is talking to, for example to look up a customer's own orders. Signed-in users explains how: your website tells the chatbot who is signed in, and in a chat app, a customer links their account once.

10. Watch it and improve it

Once your chatbot is live, the best way to make it better is to read what people ask it.

• Conversations shows every conversation, from every channel. You can filter for the ones it couldn't answer, the answers people rated badly, and the ones passed to your team. When you find a weak answer, press Fix this, write the answer you wanted, and the chatbot uses it from then on.
• Analytics shows how many conversations you get each day and how quickly your chatbot answers.
• Evaluate is for measuring instead of guessing. You write a set of questions with the answers you expect, and ChatFrom scores how well your chatbot does. Run it again after a change to see whether it helped. These runs are paid with your own OpenRouter key, not from your plan's credits.

The Overview page brings it all together: a checklist of these steps that ticks itself off as you go, your numbers for the week, and anything that needs your attention. If you get stuck, write to us.