Sooner or later every free AI tool says something like “You've reached your limit. Try again later.” To understand why, you need one new word: token.
What is a token?
AI doesn't read letters or whole words. It reads tokens: small pieces of words. Common words are usually one token. Rare or long words get cut into several pieces.
What you type
Build me a website for our Erasmus party!
What the AI sees (an example: every model cuts a bit differently)
BuildmeawebsiteforourErasmusparty!
= 10 tokens · 41 characters · 8 words. The rare word “Erasmus” became two pieces.
≈ 17 tokens · 66 characters · 13 words
Rough rule for English: 1 token ≈ 4 characters ≈ ¾ of a word. Other languages, like Turkish or Greek, often need more tokens for the same sentence.
Every time you use an AI, both your message and the AI's answer count. When you ask it to change your website, the whole page is usually sent along too:
🧾 One request to an AI (rough example)
Your prompt
40 tokens
Your web page, sent along
3,000 tokens
The AI's answer
2,000 tokens
Total
≈ 5,040 tokens
So long pages and long answers use up a free plan much faster than short questions.
Two kinds of limits
⏱️
Rate limit: how fast
How many requests you may send per second or per minute. Ask too fast and you'll see “Please wait a moment”.
📅
Usage limit: how much
How much you may use in total per day or per month: messages, tokens or credit. Used it all? “Come back tomorrow”, or next month.
It works a lot like your mobile data plan:
📱 Your mobile data plan
🤖 A free AI plan
10 GB per month
A set amount of messages, tokens or credit per day or month
Videos use your data fast
Long pages and long answers use tokens fast
When the network is busy, it gets slower
Rate limit: only a few requests per second or minute
It resets on a fixed day each month
It resets every day or every month
Out of data? Wait, or buy more
Out of tokens? Wait, switch tools, or pay
Why do companies limit free AI?
🖥️
Expensive computers
AI runs on special chips called GPUs, in huge buildings full of computers (data centres). They cost a lot of money.
⚡
Energy
Every answer uses electricity, and data centres also need cooling. Longer answers use more.
⚖️
Fairness
Millions of people share the same free computers. Limits stop one person from using everything.
And to be honest: a free plan is also a free sample. Companies hope you'll like it and pay later.
A real example: DigiBot DigiBot runs on the workshop organizer's free Mistral account. That account gets $10 of AI credit per month (it resets on the 1st). The model that writes your code can answer only about 2 requests per second, shared by the whole class. That's why DigiBot Studio asks you to wait a few seconds between requests, and sometimes shows a short queue. Everyone gets a fair turn.
Want your own allowance? Anyone can create a free Mistral account: you need an e-mail address (sometimes also a phone number for an SMS check), but no credit card. Each account gets its own monthly credit (right now $10), separate from the class. If everyone does this, nobody runs out because of someone else.
Paste your own API key (a secret code that lets an app use the AI on your account) into DigiBot Studio's settings, and DigiBot uses your credit. The key stays in your browser and is never sent to our server. Treat it like a password: never share it or paste it into an AI chat.
You're chatting with a free AI and it says: “You've reached your limit. Try again later.” What happened?
Open vs closed models
Every AI tool has a model inside: the “brain”. A model is a giant file of numbers, called weights, that the AI learned from huge amounts of text and code. The big question is: who can get that file?
🔓
Open-weight models
The company publishes the weights. Anyone can download the model, run it on their own computer or server, and even adapt it.
Can run on your own computer: private, even offline
Free to download, and it can't be taken away from you
Anyone can study, test and improve it
The biggest ones need very powerful computers
Small ones that fit on a laptop are weaker at code
🔒
Closed models
The company keeps the weights secret. You can only use the model through their app or their API.
Examples: the main models inside ChatGPT, Gemini and Claude
Often the most powerful models
Easy: just open the app
What you type goes to the company's servers
Prices, limits and rules can change at any time
No internet, no AI
“Open-weight” is not always the same as “open-source”: the training data usually stays secret. And each model has a licence, which sets rules for how you may use it.
What does DigiBot use? DigiBot is powered by AI models from Mistral AI, a European company based in France. Codestral, Mistral's code specialist, writes and edits your pages. Ministral, a small open-weight model, explains code and answers your questions.
Surprise: Codestral is not open-weight. Mistral keeps its weights private. So one company can make both open and closed models! DigiBot reaches both models through Mistral's API (a way for one program to talk to another), with data stored in EU data centres. Why this choice? They are strong at code, they come from Europe, and there is a free plan we can share with the whole class.
What does “open-weight” mean?
Which tool is right for me?
Answer three quick questions and get a suggestion. There's no wrong answer, and you can change your mind at any time.
Answer all 3 questions to see your match.
Four ways to build with AI
The tools above fit together into four typical ways of working. None is “the best”. Each one fits a different moment.
Comparison of four ways to build a website with AI
Way of working
How it works
What you control
What's free
Best for
💬 Chat app + copy/pasteLe Chat, ChatGPT, Gemini…
You ask in a chat, copy the code into a file on your computer, then open it in your browser.
Everything, but by hand. It's easy to lose track of which version is which.
Free chat plans with daily message limits.
Questions, explanations, small pieces of code.
🧑💻 AI code editorVS Code + Copilot Free, Cursor Hobby
The AI reads and edits the real files in your project folder.
A lot: you see each change and accept or reject it.
The editor is free; AI help has a monthly limit.
Bigger projects, and learning how developers really work.
⚡ AI app builderBolt.new, v0, Lovable
You describe an app; the builder writes all the code and shows a live preview.
Less: it changes lots of code at once, so it's harder to follow.
Small free credits that run out fast.
Quick prototypes and testing an idea.
🤖 DigiBot Studiothis workshop
You describe your page; DigiBot builds one HTML file with a live preview. Then you improve it step by step, or pick one part and change only that.
A lot: small steps, and DigiBot can explain any code it wrote.
Free for this class (the organizer's free Mistral credit), or add your own free Mistral key.
Learning the Plan → Prompt → Review → Fix loop and building your first site today.
On a small screen, swipe the table sideways to see all columns.
Protect your privacy
Free tools can learn from what you type Many free AI services may use your chats to train future models, and some let human reviewers read them. Anything you paste may be stored on the company's servers.
Look for a setting such as “Improve the model for everyone” or “Use my data for training”, usually under Settings → Privacy or Data controls, and switch it off.
The same goes for DigiBot: it also runs on a free AI plan. Read our privacy page to see what happens to your prompts.
✅ Fine to paste
The code of your web page
Texts for your site (with no private details)
Error messages from the browser
Your plan and ideas for the site
❌ Never paste
Passwords and API keys
Your address, phone number or ID numbers
Health, bank or other sensitive details
Other people's private information or photos
Try one tool now
Pick any chat assistant from the toolbox, open it, and paste this prompt. Then compare its answer with what you learned in Chapter 1.
Prompt for any chat assistant
Explain HTML, CSS and JavaScript to a complete beginner, using a house as an example. Use 5 short sentences.
Your turn You now know your tools, their limits, and how to stay safe. But a tool is only as good as the instructions you give it. Next: make a plan before you write a single prompt.