Skip to main content
Version: 4.xx.xx


@refinedev/inferencer is a package that provides a way to automatically generate views for resources based on the data structure. The aim is to reduce the amount of time spent on creating views for resources by generating the code automatically that can be customized easily.

The package exports components for List, Show, Create and Edit views inside UI package scopes. For example, @refinedev/inferencer/antd exports components for @refinedev/antd package.


npm i @refinedev/inferencer

Available UI Inferencers​

Good to know:

@refinedev/inferencer is an experimental package and it is now in the early stages of development. We are working on improving the package and adding new features.

If you have any suggestions or feedback, please let us know in the GitHub Discussions

@refinedev/inferencer components are meant to be used in development environments. They are not meant to be used in production environments.

How it works?​

Simply, @refinedev/inferencer generates views and codes based on the data structure of the resource by fetching it using the dataProvider of <Refine/> component.

How the data is obtained?​

For, edit and show actions, we send the request with resource and id. For list and create actions, we send a list request with resource and use one of the items to generate the view. These actions will take place in your app.

How the fields are inferred?​

While inferring the field types, we use a set of functions that each checks the field for a specific type and returns the inferred type. These functions also can return a priority field that is used to determine the type of the field. For example, if we have a created_at property with a string value, we can infer it as a date type and a text type. In this case, we use the priority field to determine the type of the field. The higher the priority, the more accurate the type of the field.

Properties with multiple values are identified as array type but also repeats the same process for their values to determine the type of the values. Same also happens for object type properties. Both can have accessor field in return value to access the values of the property which is used when creating the view and the code.

If the property is an object type, we try to pick a key to represent that property. For example, if we have a category field with { label: string; id: string; } type, we pick label as the key to represent the property. These object fields with keys to represent them have the property fieldable set to true in the return value.

Available field types and functions​

type Types =
| "relation"
| "array"
| "object"
| "date"
| "email"
| "image"
| "url"
| "richtext"
| "text"
| "number"
| "boolean"
| "unknown"
| `custom_${string}`;

custom_${string} is used by the inferencer components of UI packages when they have custom representations, for now users can't pass custom types and functions to the inferencer components.

Keys to represent object type properties​

type PresentationalKeys =
| "name"
| "label"
| "title"
| "count"
| "content"
| "username"
| "nickname"
| "login"
| "firstName"
| "lastName"
| "url";

How the relations are determined?​

There are some conditions we look for before determining if a field is can be a relation. These won't trigger any API calls to the resources.

  • If the property name ends with id or ids. camelCase, PascalCase, snake_case, kebab-case, UPPER_CASE, lower_case are all supported with or without array brackets([]).
  • If the property is an object with a single property id.
  • If the property is an array of objects with a single property id or UUID compatible strings or numbers.
  • If the property is a string or number and the property name matches with one of the known resources (singular or plural).

If one of these conditions is met, we consider the property as a relation type and try to determine the related resource.

To determine the relations;

  • First, we try to find a resource that matches with the property name (singular or plural).
  • If a resource is found in the resources array with a match, we use that resource as the related resource.
  • If no resource is found, we send two requests to the default dataProvider one with singular property name and one with plural property name, both stripped from the id suffixes if there are any.
  • If a resource is found we use that resource and its dataProvider (if specified) and make the API call with the property value.
  • If any of these requests succeed with 200 status code, we consider the property as a relation type and set the resource as the related resource.
  • If none of these requests succeed, we remove the relation mark from the property and consider it as a normal field. If it's an object type, then we will try to find the best suitable property to represent it.
Manually setting relations and resources:

If your dataProvider and resources has a different way of work that makes it impossible for Inferencer to find the relation resources. You can manually modify the inferred fields by using the fieldTransformer function. You can find more information about it in the Modifying the inferred fields section.

How the components are rendered and the code is generated?​

To render the components we use a fork of react-live package with Typescript support.

After the fields are determined, we use the renderer functions to create the code for the components and also use the same code to render the components in the view. renderer functions are constructed per action type and the UI package. This means, @refinedev/inferencer/antd and other UI scopes has different renderer functions for list, show, edit and create actions.

renderer function returns a string that includes the code for the component which is presented to user to copy and paste to their project. The same code is also used to render the component in the view.

Component name is determined by the active resource element and the active action. If the resource has option.label field, it will be used as the part of the component name. Otherwise, the will be used. For example, if the resource name is categories and the action is list, the component name will be CategoryList.

Usage with GraphQL backends and meta values​

Refine handles the GraphQL backends by using the meta properties in its data hooks. Inferencer lets you define meta values for your resources and methods in a single prop and uses it when generating the code and inferring the fields. Unlike the meta property of the data hooks, Inferencer components uses the meta property with a nested structure, letting you define the meta values per resource and action.

Here's the syntax for defining the meta values in Inferencer components:

[resourceNameOrIdentifier: string]: {
[methodName: "default" | "getList" | "getMany" | "getOne" | "update"]: Record<string, unknown>,

default is the default meta value for all the methods. In the absence of a specific meta value for a method for a resource, the default value will be used.

This structure is designed to let users provide meta values for multiple resources and actions at once since the Inferencer might find relations and try to use hooks to fetch the necessary data.

Example Usage​

posts: {
getList: {
fields: ["id", "title", "content", "category_id", "created_at"],
categories: {
default: {
fields: ["id", "title"],

Modifying the inferred fields​

If you want to customize the output of the Inferencer such as setting a custom accessor property for object type fields or changing the type of a field, or changing the resource for a relation type, you can usefieldTransformer prop in Inferencer components. It is a function that takes the field as an argument and returns the modified field. If undefined | false | null is returned, the field will be removed from the output, both for the preview and the code.

Hiding the Code Viewer and Development Warning​

If you want to hide the code viewer and the warning components, you can use the hideCodeViewerInProduction prop. This will only work in production mode. In development mode, the code viewer and the information block will always be visible.

Please note that the Inferencer components are not meant to be used in production. They are meant to be used in development mode to help you generate the code for your components.