Implementation:Langgenius Dify Pipeline Graph Configuration

Knowledge Sources	Dify
Domains	RAG, Pipeline, Frontend
Last Updated	2026-02-12 00:00 GMT

Overview

Description

Pipeline Graph Configuration documents the type system and data structures that define configurable parameters for RAG pipeline graph nodes in Dify. The core type is RAGPipelineVariable, which provides a schema-driven approach to declaring what inputs each node in the pipeline graph accepts. This type system bridges the backend's parameter definitions with the frontend's dynamic form rendering.

The implementation includes:

• RAGPipelineVariable -- The primary type defining a single configurable parameter within a pipeline node. Each variable specifies its owning node, input type, label, validation rules, and UI hints.
• PipelineInputVarType -- An enum of supported variable types that map to specific form field controls.
• VAR_TYPE_MAP -- A mapping from PipelineInputVarType to BaseFieldType, translating backend variable types to frontend form field implementations.
• ChunkingMode -- An enum defining the three supported document chunking strategies.

Usage

These types are used throughout the pipeline editor and configuration panels to:

1. Define the shape of parameter responses from the processing and pre-processing parameter endpoints.
2. Drive dynamic form generation in node configuration panels.
3. Validate user inputs against the declared constraints (required, max_length, options).
4. Associate parameters with their owning nodes in multi-node pipeline graphs.

Code Reference

Source Location

web/models/pipeline.ts, lines 116--172

Signature

export enum PipelineInputVarType {
  textInput = 'text-input',
  paragraph = 'paragraph',
  select = 'select',
  number = 'number',
  singleFile = 'file',
  multiFiles = 'file-list',
  checkbox = 'checkbox',
}

export const VAR_TYPE_MAP: Record<PipelineInputVarType, BaseFieldType> = {
  [PipelineInputVarType.textInput]: BaseFieldType.textInput,
  [PipelineInputVarType.paragraph]: BaseFieldType.paragraph,
  [PipelineInputVarType.select]: BaseFieldType.select,
  [PipelineInputVarType.singleFile]: BaseFieldType.file,
  [PipelineInputVarType.multiFiles]: BaseFieldType.fileList,
  [PipelineInputVarType.number]: BaseFieldType.numberInput,
  [PipelineInputVarType.checkbox]: BaseFieldType.checkbox,
}

export type RAGPipelineVariable = {
  belong_to_node_id: string
  type: PipelineInputVarType
  label: string
  variable: string
  max_length?: number
  default_value?: string
  placeholder?: string
  unit?: string
  required: boolean
  tooltips?: string
  options?: string[]
  allowed_file_upload_methods?: TransferMethod[]
  allowed_file_types?: SupportUploadFileTypes[]
  allowed_file_extensions?: string[]
}

export type InputVar = Omit<RAGPipelineVariable, 'belong_to_node_id'>
export type RAGPipelineVariables = RAGPipelineVariable[]

The ChunkingMode enum (defined in web/models/datasets.ts):

export enum ChunkingMode {
  text = 'text_model',       // General text chunking
  qa = 'qa_model',           // QA pair chunking
  parentChild = 'hierarchical_model', // Parent-Child hierarchical chunking
}

Import

import type {
  RAGPipelineVariable,
  RAGPipelineVariables,
  InputVar,
  PipelineInputVarType,
} from '@/models/pipeline'
import { VAR_TYPE_MAP } from '@/models/pipeline'
import { ChunkingMode } from '@/models/datasets'

I/O Contract

RAGPipelineVariable Fields

Field	Type	Required	Description
belong_to_node_id	string	Yes	ID of the owning node, or 'shared' for pipeline-level variables
type	PipelineInputVarType	Yes	Input type: text-input, paragraph, select, number, file, file-list, or checkbox
label	string	Yes	Human-readable label for display
variable	string	Yes	Machine-readable variable name used in API requests
max_length	number	No	Maximum character length for text inputs
default_value	string	No	Default value pre-populated in the form
placeholder	string	No	Placeholder text for empty input fields
unit	string	No	Unit label for numeric inputs (e.g., "tokens", "characters")
required	boolean	Yes	Whether the parameter is mandatory
tooltips	string	No	Help text shown on hover
options	string[]	No	Available choices for select-type variables
allowed_file_upload_methods	TransferMethod[]	No	Permitted file upload methods for file-type variables
allowed_file_types	SupportUploadFileTypes[]	No	Allowed MIME types for file uploads
allowed_file_extensions	string[]	No	Allowed file extensions for file uploads

PipelineInputVarType to BaseFieldType Mapping

PipelineInputVarType	BaseFieldType	Form Control
text-input	textInput	Single-line text field
paragraph	paragraph	Multi-line textarea
select	select	Dropdown selector
number	numberInput	Numeric input with optional unit
file	file	Single file upload
file-list	fileList	Multiple file upload
checkbox	checkbox	Boolean checkbox

Usage Examples

import type { RAGPipelineVariable } from '@/models/pipeline'
import { VAR_TYPE_MAP, PipelineInputVarType } from '@/models/pipeline'
import { BaseFieldType } from '@/app/components/base/form/form-scenarios/base/types'

// Map a pipeline variable to a form field type
function getFieldType(variable: RAGPipelineVariable): BaseFieldType {
  return VAR_TYPE_MAP[variable.type]
}

// Render dynamic form fields from pipeline variables
function renderNodeConfig(variables: RAGPipelineVariable[], nodeId: string) {
  const nodeVars = variables.filter(v => v.belong_to_node_id === nodeId)
  return nodeVars.map(v => ({
    fieldType: VAR_TYPE_MAP[v.type],
    label: v.label,
    name: v.variable,
    required: v.required,
    defaultValue: v.default_value,
    placeholder: v.placeholder,
    options: v.options,
    maxLength: v.max_length,
  }))
}

// Check chunking mode from a pipeline template
import { ChunkingMode } from '@/models/datasets'

function isHierarchicalChunking(mode: ChunkingMode): boolean {
  return mode === ChunkingMode.parentChild
}

Related Pages

• Principle:Langgenius_Dify_Embedding_Indexing_Configuration