Build Your Backend with AI Assistants That Follow One Standard

Intro

0:00 An AI assistant writes good code when it knows the platform, sees the project, and gets checked. In this video : the docs written for assistants, your backend as files with types, and the checks that hold for code whoever wrote it.

AI on a backend, the usual way

0:15 An assistant on a backend usually invents methods the framework never had, guesses the names of columns, never sees the rules that keep data private, and leaves the review to you. With API Maker, the docs come in plain text for assistants, every table and every call is typed in the project, the security report sits in the repository, and the same checks run on every line, written by a person or a model.

0:42 In this video : the docs for assistants, llms dot txt, prompts that work, the security report as a file, the types of your tables, and a mistake caught before it runs.

The docs for assistants

0:54 The docs have a page for AI assistants. Every page has this bar : copy it as Markdown, view the Markdown, or open it in ChatGPT or Claude with the page loaded. And the files an assistant can read : an index of every page, the whole documentation in one file, and the Markdown of any page.

llms.txt

1:15 Any page as plain Markdown, without the navigation. And llms dot txt : one line per page, its title, what it covers and the address of its Markdown, grouped like the navigation, so an agent fetches only the pages it needs.

Prompts that work

1:31 Point the assistant at the pages it needs, then ask : write a custom API that returns the ten latest orders of the signed-in person. It reads how custom APIs and g work, and writes code the platform runs.

The security report as a file

1:45 In the project, the repository of the account. In src, the API security report : the score, the findings by severity, and for each one, the evidence and what to do. An assistant reads it, fixes the hooks, groups and settings it names, and the next scan says if it worked.

The types of your tables

2:04 Next to it, types : db interfaces, generated from the schemas of every database. The coupons of MariaDB : code, campaign, discount pct, valid until.

A mistake caught before it runs

2:17 Back in Cart Quote, the discount reads the column discount pct. Say the code reads discount percent instead, a name that sounds right. The editor flags it at once : this property does not exist on the coupons ; did you mean discount pct? Whoever wrote the line, a person or a model, it never reaches a customer. So the loop is short. The assistant writes in the folder. The types check every table and every call.

2:48 The local client syncs what it saves. npm run test runs the tests, and the security report says what is still open.

Built on standards

2:57 Every account has the same layout, so an assistant finds the custom APIs, schemas and settings in the same places. Every rule is a file it can read. And nothing it writes skips the types, the tests or the scan.

Recap

3:13 To sum up. Docs in plain text for assistants. Your backend as files, with the types of every table. The security report in the repository. The same checks for every line, whoever wrote it. Next : Git integration. API Maker. Build faster, spend smarter, deliver better.