Skip to content

overview.md

Latest commit

 

History

History
327 lines (259 loc) · 8.98 KB

overview.md

File metadata and controls

327 lines (259 loc) · 8.98 KB

Workflow Overview

A Workflow in AgenticFlow is a linear, sequential automation pipeline that executes nodes step-by-step from top to bottom.

Core Concepts

Every workflow consists of three main components:

Component Description Schema Field
Input Schema Parameters users provide when starting the workflow input_schema
Nodes Sequential processing steps that execute tasks nodes
Output Mapping Which node outputs to return as final results output_mapping

Execution Model

  • Linear execution: Nodes run one-by-one, top to bottom
  • No branching: Every node executes every time
  • No loops: Nodes run exactly once in sequence
  • Forward data flow: Later nodes can reference earlier nodes' outputs

Input Schema

Defines what data the workflow needs before execution.

Supported Input Types

Type UI Type Use Case
Text short_text Single-line text input
Long Text long_text Multi-line text input
Number number Numeric values
Checkbox checkbox Boolean true/false
Select select Dropdown single-select
Multi-Select multi_select Dropdown multi-select
Image Upload media_url Image file upload
Video Upload media_url Video file upload
Audio Upload media_url Audio file upload
File to URL file_url Any file upload

Input Schema Example

{
  "input_schema": {
    "type": "object",
    "title": "User inputs",
    "description": "User inputs for the workflow",
    "required": ["image_to_url", "name", "description"],
    "properties": {
      "image_to_url": {
        "type": "string",
        "title": "Reference Image",
        "description": "",
        "ui_metadata": {
          "type": "media_url",
          "media_type": "image",
          "order": 0,
          "placeholder": "",
          "value": "",
          "allowed_mime_types": ["image/jpeg", "image/png", "image/gif"],
          "max_size": 26214400
        }
      },
      "name": {
        "type": "string",
        "title": "Workflow Name",
        "description": "",
        "ui_metadata": {
          "type": "short_text",
          "order": 1,
          "value": "",
          "placeholder": "Enter name..."
        }
      },
      "description": {
        "type": "string",
        "title": "Description",
        "description": "",
        "ui_metadata": {
          "type": "short_text",
          "order": 2,
          "value": "",
          "placeholder": "Enter description..."
        }
      }
    }
  }
}

Input Field Properties

Field Description
type Data type: string, number, boolean, array
title Display label in UI
description Help text (optional)
ui_metadata.type UI input type (see Supported Input Types)
ui_metadata.order Display order (0, 1, 2...)
ui_metadata.value Default value
ui_metadata.placeholder Placeholder text

Media URL Options

For media_url type inputs:

Field Description
media_type image, video, or audio
allowed_mime_types Array of allowed MIME types
max_size Max file size in bytes (default: 26214400 = 25MB)

Nodes

Nodes are the building blocks that perform actions in a workflow. Each node is created from a node_type selected from the available node types list.

Node Structure

Each node in the nodes array has:

{
  "name": "unique_node_name",
  "title": "Human-readable Title",
  "description": "What this node does",
  "node_type_name": "selected_node_type",  // From node_type list
  "input_config": {
    // Configuration based on node_type's input schema
  },
  "connection": "{{__app_connections__['connection-id']}}",
  "output_mapping": null
}

How Nodes Work

  1. Select a node_type from the available list (e.g., claude_ask, generate_image)
  2. Configure input_config based on that node_type's required fields
  3. Connect to services if the node_type requires external API access
  4. Reference data from inputs or previous nodes using {{...}} syntax

Node Type Categories

Node types are organized by category:

Category Description Example node_types
AI/LLM AI model calls, text generation claude_ask, openai_chat, gemini
Image Generation Create images from prompts generate_image, dall_e
Data Processing Transform and manipulate data json_parse, text_transform
Integrations Connect to external services (300+ MCPs) slack_send, gmail, notion
API Calls HTTP requests, webhooks http_request, webhook
File Operations Upload, download, process files file_upload, pdf_parse

Note: For the complete list of available node_types and their configurations, see Node Types Reference

Example: LLM Node

{
  "name": "claude_ask",
  "title": "Ask Claude",
  "node_type_name": "claude_ask",
  "input_config": {
    "model": "claude-3-haiku-20240307",
    "prompt": "Generate a description for {{name}}: {{description}}",
    "max_tokens": 1000,
    "temperature": 0.5,
    "system_message": "You are a helpful assistant."
  },
  "connection": "{{__app_connections__['connection-id']}}"
}

Example: Image Generation Node

{
  "name": "generate_image_1",
  "title": "Generate Image",
  "node_type_name": "generate_image",
  "input_config": {
    "prompt": "Generate thumbnail for {{name}} in purple theme like {{image_to_url}}",
    "provider": "Nano Banana Pro",
    "aspect_ratio": "9:16 (1K)",
    "negative_prompt": "NSFW"
  },
  "connection": "{{__app_connections__['connection-id']}}"
}

Parameter Substitution

Use {{...}} syntax to reference data between workflow steps.

Reference Types

Syntax Description Example
{{input_name}} Reference workflow input {{name}}, {{description}}
{{node_name.field}} Reference node output field {{claude_ask.result}}
{{node_name}} Reference entire node output {{generate_image_1}}

Usage in Prompts

"prompt": "Generate a description for my workflow about {{name}} which works like this: {{description}}"

Nested Field Access

{{api_response.data.items[0].title}}

Output Mapping

Controls what the workflow returns after execution.

When Defined

{
  "output_mapping": {
    "description": "{{claude_ask.result}}",
    "thumbnail": "{{generate_image_1.image_url}}"
  }
}

When Empty

If output_mapping is empty {}, the workflow returns all outputs from the last node.

Connections

Nodes that require external services use connections:

"connection": "{{__app_connections__['connection-uuid']}}"

For available connection providers, see connections.md

Workflow Metadata

Additional workflow properties:

Field Description
id Unique workflow identifier
name Display name
description Workflow description
public_runnable Can others run this workflow
public_clone Can others clone this workflow
num_runs Total execution count
num_views Total view count

Complete Workflow Example

{
  "name": "Marketplace Asset Generator",
  "input_schema": {
    "type": "object",
    "required": ["name", "description", "image_to_url"],
    "properties": {
      "name": { "type": "string", "ui_metadata": { "type": "short_text" } },
      "description": { "type": "string", "ui_metadata": { "type": "short_text" } },
      "image_to_url": { "type": "string", "ui_metadata": { "type": "media_url" } }
    }
  },
  "nodes": {
    "nodes": [
      {
        "name": "claude_ask",
        "node_type_name": "claude_ask",
        "input_config": {
          "prompt": "Generate description for {{name}}: {{description}}"
        }
      },
      {
        "name": "generate_image_1",
        "node_type_name": "generate_image",
        "input_config": {
          "prompt": "Generate thumbnail for {{name}} workflow"
        }
      }
    ]
  },
  "output_mapping": {}
}

Web UI URLs

Page URL Pattern
Workflow List https://agenticflow.ai/app/workspaces/{workspace_id}/workflows
Workflow Details https://agenticflow.ai/app/workspaces/{workspace_id}/workflows/{workflow_id}/build
Workflow Run Details https://agenticflow.ai/app/workspaces/{workspace_id}/workflows/{workflow_id}/logs/{workflow_run_id}

Related Documentation