Workflow Application

Overview

The Workflow Application is the most powerful application type in Nadoo AI. Instead of a single LLM conversation, a Workflow App uses a visual node graph to define complex, multi-step execution flows. Nodes represent individual operations — AI model calls, conditional branches, loops, knowledge base queries, HTTP requests, code execution — and edges define the data flow between them. Workflow Apps are ideal for:

Multi-step AI agent pipelines with branching logic
Data processing and transformation workflows
Orchestrated multi-agent systems
Automated business processes with AI decision-making

How It Works

A workflow is a directed graph built in the visual editor. Execution begins at a Start Node, flows through connected nodes, and terminates when all paths reach an End Node or an unconnected output. Each node processes its input, executes its operation, and passes the result to the next node(s) via edges. Variables flow through the graph, accumulating context as execution progresses.

Node Types

Nadoo AI provides 18+ built-in node types organized by category.

Core Nodes

Node	Description
Start	Entry point of every workflow. Receives user input and initial variables.
End	Terminal node that returns the final output to the user.
AI Agent	Calls an AI model with configurable system prompt, tools, and agent strategy.
LLM	Direct LLM call without agent capabilities — simpler and faster for basic generation.

Logic Nodes

Node	Description
Condition	Branches execution based on a boolean expression or comparison.
Loop	Iterates over a list, executing child nodes for each item.
Switch	Routes execution to one of multiple paths based on a value match.
Merge	Combines outputs from parallel branches into a single output.

Data Nodes

Node	Description
Search Knowledge	Queries a knowledge base using vector, keyword, or hybrid search.
HTTP Request	Makes HTTP calls to external APIs with configurable method, headers, and body.
Code Executor	Runs custom Python or JavaScript code with access to workflow variables.
Variable Setter	Sets or updates a workflow variable.
Template	Renders a text template with variable interpolation using Jinja2 syntax.

Integration Nodes

Node	Description
Tool Call	Invokes a registered plugin or tool.
Webhook Trigger	Starts a workflow from an incoming webhook request.
Email	Sends an email with a configurable template.
Notification	Sends a notification to a configured channel (Slack, Discord, etc.).

Advanced Nodes

Node	Description
Sub-Workflow	Embeds and executes another workflow as a nested operation.
Human Review	Pauses execution and waits for human approval before continuing.
Parallel	Executes multiple branches simultaneously and waits for all to complete.

Building a Workflow

Create a Workflow App

From the workspace dashboard, click New Application and select Workflow as the type.

Open the Visual Editor

The workflow editor opens with a Start Node already placed on the canvas.

Add Nodes

Drag nodes from the node palette onto the canvas. Connect them by dragging from an output port to an input port.

Configure Each Node

Click a node to open its configuration panel. Set the model, prompt, condition expression, or other parameters specific to the node type.

Define Variables

Use the Variables panel to define workflow-level variables that nodes can read and write.

Test and Debug

Click Run to execute the workflow with test input. The visual editor highlights the active node and shows real-time output at each step.

AI Agent Node

The AI Agent node is the most commonly used node. It wraps an LLM with an agent strategy that determines how the model reasons and uses tools.

Agent Strategies

Strategy	Description
Standard	Single LLM call with optional tool use. Fastest execution.
Chain of Thought (CoT)	Model reasons step-by-step before producing a final answer.
ReAct	Interleaves reasoning and action — the model thinks, acts (calls a tool), observes the result, and repeats.
Function Calling	Model selects and invokes tools using structured function call syntax.
Reflection	Model generates an initial response, critiques it, and refines it.
Tree of Thoughts	Explores multiple reasoning paths in parallel and selects the best outcome.

ReAct and Function Calling are the most versatile strategies for workflows that involve tool use. Use Chain of Thought when you need transparent, step-by-step reasoning without tools.

Execution and Streaming

When a Workflow App executes, Nadoo AI streams progress events in real time via Server-Sent Events (SSE). This allows the frontend to display which node is running, show intermediate results, and render the final output as it is generated.

Workflow-Level Events

Event	Description
`workflow_start`	Workflow execution has begun
`workflow_end`	Workflow execution completed successfully
`workflow_error`	Workflow execution failed with an error

Node-Level Events

Event	Description
`node_start`	A specific node has started executing
`node_end`	A specific node has finished executing
`node_error`	A specific node encountered an error

Agent-Level Events

Event	Description
`agent_iteration`	An agent reasoning iteration has occurred
`agent_tool_call`	The agent is invoking a tool
`agent_tool_result`	A tool has returned a result
`agent_thinking`	The agent is in a thinking/reasoning phase
`cot_step`	A chain-of-thought reasoning step

See Real-time Streaming for the full event reference with example payloads.

Running a Workflow

Via the Chat Interface

Workflow Apps can be used through the same chat interface as Chat Apps. The user sends a message, the workflow executes, and the final output is streamed back. Intermediate steps (node execution, tool calls, reasoning) can be displayed in an expandable execution monitor.

Via the API

POST /api/v1/workflow/execute
{
  "application_id": "app-uuid",
  "inputs": {
    "query": "Summarize last quarter's sales report"
  },
  "stream": true
}

The response is an SSE stream with workflow, node, and agent events.

Via Scheduling

Automate workflow execution on a schedule using cron expressions. See Scheduling for details.

Versioning and Publishing

Workflows support versioning to manage changes safely.

Concept	Description
Draft	The working version visible in the editor. Changes are saved automatically.
Published	The live version that serves user requests. Immutable once published.
Version History	A log of all published versions with timestamps and change descriptions.

Publishing a Workflow

Click Publish in the workflow editor toolbar
Enter a version description (e.g., “Added fallback handling for API errors”)
The current draft becomes the new published version
Previous published versions remain in the version history for rollback

Publishing replaces the live version immediately. Test thoroughly in the editor before publishing. You can roll back to any previous version from the version history.

Best Practices

Start simple, then expand

Begin with a linear flow: Start -> AI Agent -> End. Add branching, loops, and tools incrementally as you validate each step.

Use descriptive node names

Rename nodes to describe their purpose (e.g., “Classify Intent” instead of “AI Agent 1”). This makes complex workflows readable at a glance.

Handle errors at every step

Add Condition nodes after critical operations (HTTP requests, tool calls) to handle failure cases gracefully. Use the node_error event to surface issues to users.

Leverage sub-workflows

Extract reusable logic into sub-workflows. This keeps the main workflow clean and makes common patterns shareable across applications.

Test with edge cases

Use the visual debugger to step through execution with various inputs — including empty inputs, malformed data, and adversarial prompts.

Next Steps

Visual Editor

Learn the workflow editor interface and keyboard shortcuts

AI Agent Node

Deep dive into the AI Agent node and agent strategies

Scheduling

Automate workflow execution with cron schedules

Channel App

Deploy your workflow to messaging platforms

Getting Started

Applications

AI Models

Knowledge Base

Knowledge Graphs

Messaging Channels

Chat & Multimodal

Workspaces & Security

Resources

Workflow Application

Overview

How It Works

Node Types

Core Nodes

Logic Nodes

Data Nodes

Integration Nodes

Advanced Nodes

Building a Workflow

AI Agent Node

Agent Strategies

Execution and Streaming

Workflow-Level Events

Node-Level Events

Agent-Level Events

Running a Workflow

Via the Chat Interface

Via the API

Via Scheduling

Versioning and Publishing

Publishing a Workflow

Best Practices

Next Steps

Visual Editor

AI Agent Node

Scheduling

Channel App

Getting Started

Applications

AI Models

Knowledge Base

Knowledge Graphs

Messaging Channels

Chat & Multimodal

Workspaces & Security

Resources

​Overview

​How It Works

​Node Types

​Core Nodes

​Logic Nodes

​Data Nodes

​Integration Nodes

​Advanced Nodes

​Building a Workflow

​AI Agent Node

​Agent Strategies

​Execution and Streaming

​Workflow-Level Events

​Node-Level Events

​Agent-Level Events

​Running a Workflow

​Via the Chat Interface

​Via the API

​Via Scheduling

​Versioning and Publishing

​Publishing a Workflow

​Best Practices

​Next Steps

Visual Editor

AI Agent Node

Scheduling

Channel App

Overview

How It Works

Node Types

Core Nodes

Logic Nodes

Data Nodes

Integration Nodes

Advanced Nodes

Building a Workflow

AI Agent Node

Agent Strategies

Execution and Streaming

Workflow-Level Events

Node-Level Events

Agent-Level Events

Running a Workflow

Via the Chat Interface

Via the API

Via Scheduling

Versioning and Publishing

Publishing a Workflow

Best Practices

Next Steps