For AI agents

An AI that doesn't know the library writes a date picker from scratch, invents a red class for errors and sorts the table on the client. Two files fix that without you explaining it again in every conversation: the full reference in plain text and three lines in your project's AGENTS.md.

llms.txt

The whole API in one text file, made to be read by a model and not by people. Point the AI to it:

https://juniorcarlini.github.io/tucano/llms.txt

First comes a section per component, with the declarative path through data-*, which is the normal one in a Django project. Then the full reference: selector, attributes, every option with the comment that explains it, methods, events, CSS classes and tokens. That part is generated from the code on every build — if an option isn't there, it doesn't exist. The rules about what not to do come at the end. The prose in the file is written in Portuguese, while the API names — attributes, options, methods, events — are in English, so a model reads it without trouble.

Every page on this site advertises the file in its <head>, for tools that look for it:

<link rel="alternate" type="text/plain" href="https://juniorcarlini.github.io/tucano/llms.txt">

AGENTS.md in your project

Paste this into the AGENTS.md at the root of the project that uses Tucano — or into CLAUDE.md, .cursorrules, wherever your tool looks. It's what makes the AI pick the library instead of writing a component from scratch.

## Interface
This project uses Tucano. Do not write from scratch a date picker, searchable select, color picker, upload, mask, toast, tooltip, modal, drawer, accordion, tabs, dropdown menu, table, pagination or text editor.
Full reference: https://juniorcarlini.github.io/tucano/llms.txt

The list names the components that have JavaScript, which are the ones an AI rewrites most. The pieces that are only classes — button, alert, badge, loading, timeline, checkbox, radio and switch, side menu, label and form field — are in the same llms.txt, and the rule "don't design a second button" covers the rest.

Three lines are enough because the work is in the link. If you want the AI to get it right in the first answer, without opening the reference, add the examples your project uses most:

## Interface
This project uses Tucano. Do not write from scratch a date picker, searchable select, color picker, upload, mask, toast, tooltip, modal, drawer, accordion, tabs, dropdown menu, table, pagination or text editor.
Full reference: https://juniorcarlini.github.io/tucano/llms.txt

<input data-tuc-datepicker data-mode="range">
<select data-tuc-select multiple>
<input data-tuc-mask="cpf-cnpj">
<input type="file" data-tuc-upload>
<button data-tuc-tip="..." class="tuc-btn is-primary">

Tucano.toast.success('Saved');
await Tucano.confirm({ title: 'Delete?' });
Tucano.drawer({ title: 'Filters', side: 'right' });

The native element still owns the value: name, required and Django's getlist() keep working. Do not replace it with JS-only state.

And Tucano's own AGENTS.md

The repository's AGENTS.md is for people who will work on the code of the library itself: how to run it, the structure and the decisions that shouldn't be reverted without a reason, each with the bug that prompted it. To use Tucano in a project, you don't need it — you need llms.txt.

What the AI must not do

These are the rules that close llms.txt. Each one exists because it's the mistake a model makes when it guesses instead of reading.