API Maker
The backend engine for the AI era
AuthorizationEpisode 24 6:54 12 chapters November 2026
Who may call which API, read which column and write what: set once, checked on every call, with no line of auth code. The support desk of Calm Home gets its own key: public, token or no access per API, an API user with its password in the secret, groups that name tables, APIs and columns, and a token that only reads.
In the playlists The complete course Security & access
Who can call which API? Who can read which column, and write what? In API Maker, you set this once, and it is checked on every call, with no auth code to write. In this video, the support desk of Calm Home gets its own key.
Authorization is usually hard to get right. Middleware is written route by route. Role checks are spread through the handlers. A new endpoint is left open by mistake. And every model is copied, just to hide a column. With API Maker, every API has an access type. Each application is an API user, with a token. Groups name the tables, APIs, and columns it can use. And all of it is checked before any code runs.
In this video: access types, the two checks of a call, a group, columns, an API user and its token, a call refused, then allowed, and a token that can only read.
Every API has one of three access types. Public: anyone can call it, with no token. Token, the default: the caller needs a token, and a group that allows this API. No access: nothing from outside can call it, only the admin panel and your own code. You can set it on an API, a table, or a database. The closest setting wins. Open Instance API. Pick MongoDB, then catalog, and its categories.
Click Add Collection Settings. TOKEN_ACCESS is the default. Pick IS_PUBLIC, and Save. From outside, with no token, we get the eight categories. The customers still need a token. So we get four hundred and one: the header x-am-authorization is not valid. Now pick NO_ACCESS, and Save. From outside, we get four hundred and one: you are not allowed to call this API. But the admin panel and the code inside API Maker can still reach it.
So this table is for your own code only.
A call with token access passes two checks. First, the token, in x-am-authorization. It names an API user, which is an application. Then, its groups. One of them must allow this API, by instance, database, table, and API. And it can read only the columns that the group lists.
Open Security, then API Group Permissions. The group Default allows everything. Click Add New, name it Support team, and click Save Group. Now select it, and you see what it can call. Instance APIs come first. Pick the instance postgres, the database crm, its customers, and its support tickets. Just two tables. A table added tomorrow stays closed to this group. Each table lists its APIs. Pick Get all on the customers, Get all on the tickets, and Get by id.
No save, no update, and no remove. This group only reads.
Open Schema Fields to see the columns of each table. We tick the id, first name, last name, the city, status, and loyalty tier, each with R, for read. Email and phone are left out. Click Save group settings.
Open API User Permissions. There is one API user for each application. Click Add New, name it Support Desk, with the username support_desk. Its password is read from a key of the default secret, so it never goes to Git. Click Save User. For now, it is in no group. The application asks the token API for a token, with its username and password. The token works for three days.
Inside it are its username, the time of its last password change, and an end date. A new password ends every older token. With this token, we get a customer, without email and phone. But the orders of MySQL give four hundred and three.
We see the same in API Testing. Open API Testing. Pick postgres, then crm, then customers, and Get all. Set x-am-authorization to Support Desk. Send. The answer is four hundred and three. Support Desk has no access to this API, because it is in no group yet. Open API User Permissions, and pick Support Desk. Tick the group Support team, and Save. Back in API Testing, still as Support Desk, Send.
Now we get the customers, with their names, city, status, and tier. There is no email and no phone, only the columns of the group.
Now we try a write, with Save single-multiple, as Support Desk. In the body, we put a new customer. The answer is four hundred and three, because the group only reads. A read only token is just that: an API user in groups that only read.
Good to know. Calls made from code with g.sys.db do not pass these checks. Your code is trusted, but the outside is not. Column rules follow the groups of the API user, which is the application. Each API user can publish Swagger docs of exactly the APIs it can call. Groups, API users, and settings are files in Git, reviewed like code.
A group opens only what it names. A new table or API stays closed to it. No check lives in the code. So code written by a person or an AI assistant cannot forget one. And one place answers who can call what.
To sum up. Public, token, or no access, for each API, table, or database. One API user for each application, with its password in the secret. Groups that name tables, APIs, and columns. Refused until allowed, and read only when you want it. Next: auth providers, the tokens of the people who use your apps. API Maker. Build faster, spend smarter, deliver better.
Length 8:30Auth providers
Length 6:17Roles
Length 12:24Security report
Length 9:18Code scan