Skip to content

Agent Workflows

The Agent Workflows feature provides a visual workflow builder that allows users to create, configure, and manage multi-agent workflows using an intuitive drag-and-drop canvas interface. This enables the orchestration of multiple agents with conditional logic, input handling, and output management.

Overview

Agent Workflows allow you to visually design complex workflows by connecting multiple agents together. Instead of manually coordinating agent calls, you can create a workflow that automatically routes data between agents based on your defined flow.

Key Capabilities:

  • Visual Workflow Design: Drag-and-drop interface for building agent workflows
  • Multi-Agent Orchestration: Connect multiple agents in sequence or parallel
  • Conditional Branching: Route data based on conditions and agent outputs
  • Input/Output Management: Define workflow inputs and structure final outputs
  • Reusable Workflows: Save and reuse workflow configurations

Workflow Components

Canvas

The canvas is the main workspace where you design your workflow. It provides:

  • Drag-and-Drop: Place nodes by dragging them from the sidebar onto the canvas
  • Pan & Zoom: Navigate large workflows with zoom controls and canvas panning
  • Grid Alignment: Nodes snap to a grid for clean visual organization
  • Canvas Controls: Zoom in/out, fit view, and clear canvas buttons

Node Types

Workflows are built using four types of nodes:

1. Input Node

The Input Node defines the entry point of your workflow and specifies what data the workflow expects to receive.

  • Defines a single input key: query
  • Supports different data types: string, integer, JSON
  • Only one Input node is allowed per workflow
  • Connects to Agent or Condition nodes

Configuration:

Property Description
Node Name Custom name for the input node
Input Keys List of expected input parameters with their types

2. Agent Node

The Agent Node represents an agent that will be executed as part of the workflow.

  • Select from available agents in IAF
  • Configure which inputs the agent can access
  • Multiple Agent nodes can be added to a workflow
  • Receives data from Input, other Agents

Configuration:

Property Description
Node Name Custom name for the agent node
Agent Select an agent from the dropdown
Accessible Inputs Choose which workflow inputs and agent outputs this agent can access

Accessible Inputs Options:

  • All Inputs: Agent has access to all workflow inputs and outputs from other agents
  • Specific Inputs: Select individual inputs (from Input node) and agent outputs to make available

3. Condition Node

The Condition Node enables conditional branching in your workflow based on expressions.

  • Define conditions to evaluate agent outputs
  • Route data to different paths based on results
  • Supports multiple output connections for branching logic
  • Useful for error handling, validation, or decision trees

Configuration:

Property Description
Node Name Custom name for the condition node
Condition Condition to evaluate, written in plain English (e.g., "If the status of result is equal to success, then...")

4. Output Node

The Output Node defines the final output of your workflow.

  • Specifies the output schema for the workflow
  • Multiple Output nodes are allowed for different output paths
  • Receives data from Agent or Condition nodes

Configuration:

Property Description
Node Name Custom name for the output node
Output Schema JSON schema defining the structure of the output (optional)

Connections

Connections (edges) define the flow of data between nodes in your workflow.

Creating Connections

  1. Click on the output point of a source node
  2. Drag to the input point of a target node
  3. Release to create the connection

Connection Rules

From ↓ / To → Input Agent Condition Output
Input ❌ ✅ ✅ ✅
Agent ❌ ✅ ✅ ✅
Condition ❌ ✅ ❌ ✅
Output ❌ ❌ ❌ ❌
  • Input nodes can connect to Agent, Condition, or Output nodes
  • Agent nodes can connect to other Agents, Conditions, or Output nodes
  • Condition nodes can connect to Agent or Output nodes (not to other Conditions)
  • Output nodes cannot have outgoing connections

Properties Panel

When you click on a node, the Properties Panel opens on the side, allowing you to configure the selected node.

The panel displays:

  • Node Name: Editable name for the node
  • Node Type: The type of node (Input, Agent, Condition, Output)
  • Node ID: Auto-generated unique identifier
  • Type-specific settings: Configuration options based on node type

Saving a Workflow

To save your workflow:

  1. Ensure you have at least one Input node and one Agent node with a selected agent
  2. Click the Save button
  3. Enter a Workflow Name (required)
  4. Add an optional Description
  5. Click Save to store the workflow

Save Requirements:

  • At least one Input node
  • At least one Agent node with a selected agent
  • Workflow name is required

Example Workflow

Here's an example of a simple research workflow:

┌─────────┐     ┌──────────────┐     ┌─────────────┐     ┌────────┐
│  Input  │ ──► │ Research     │ ──► │ Writer      │ ──► │ Output │
│  Node   │     │ Agent        │     │ Agent       │     │ Node   │
└─────────┘     └──────────────┘     └─────────────┘     └────────┘
  1. Input Node: Receives query (string) and context (JSON)
  2. Research Agent: Searches for information based on the query
  3. Writer Agent: Takes research output and generates a formatted response
  4. Output Node: Returns the final structured response

Using Workflows in Chat Inference

In the Chat Inference tab, you can select a workflow as the agent type and interact with it just like any other agent template. This allows you to test and validate your multi-agent workflows directly from the chat interface.

Best Practices

  1. Name your nodes clearly: Use descriptive names that indicate each node's purpose
  2. Start simple: Begin with a linear workflow before adding conditions
  3. Test incrementally: Save and test your workflow as you build it
  4. Use conditions wisely: Add conditional logic for error handling and validation
  5. Document your workflows: Add descriptions when saving to help others understand the workflow