Skip to content

Linear Workflow Configuration

Linear workflows provide a straightforward way to execute a series of components in a predefined, sequential order. They are perfect for tasks that follow a linear process, such as "first run the data-ingestion agent, then run the analysis agent."

A linear workflow configuration is a JSON or YAML object with a type field set to "linear_workflow".

How It Works

When you execute a linear workflow, the framework iterates through the steps list in order. The output from one step is passed as the input to the next, allowing you to chain components together to create a processing pipeline.

Schema

The WorkflowConfig defines the structure for a linear workflow.

These fields define the fundamental properties of the workflow.

Field Type Required Description
name string Yes A unique identifier for the workflow. This name is used to run the workflow from the CLI or API.
description string No A brief, human-readable description of what the workflow accomplishes.
steps list[string or object] Yes An ordered list of the components to execute. See "Steps Configuration" below for details.
include_history boolean None If set, overrides the include_history setting for all agents in the workflow, forcing them all to either save or discard history.

The core of a linear workflow is the steps list. Each item in the list can be either a simple string (the component name) or a detailed object.

Simple Step (by Name)

The easiest way to define a step is to provide the name of the component.

"steps": ["fetch-data-agent", "process-data-agent"]

Detailed Step (by Object)

For more clarity, you can specify the component type along with its name. This is useful if you have components of different types with the same name.

"steps": [
  { "name": "fetch-data-agent", "type": "agent" },
  { "name": "analysis-pipeline", "type": "linear_workflow" }
]

The type can be "agent", "linear_workflow", or "custom_workflow".

Configuration Examples

This example defines a two-step workflow for processing customer feedback.

{
  "type": "linear_workflow",
  "name": "customer-feedback-pipeline",
  "description": "Fetches customer feedback, analyzes sentiment, and saves the result.",
  "steps": ["fetch-feedback-agent", "analyze-sentiment-agent"]
}

This workflow forces all agents within it to maintain conversation history, which is useful for debugging or creating a continuous conversational experience across multiple agents.

{
  "type": "linear_workflow",
  "name": "stateful-support-flow",
  "description": "A multi-agent support flow that remembers the conversation.",
  "include_history": true,
  "steps": ["greeting-agent", "support-agent", "followup-agent"]
}

This example shows how a linear workflow can include another workflow as one of its steps.

{
  "type": "linear_workflow",
  "name": "daily-reporting-process",
  "description": "A top-level workflow that orchestrates other workflows and agents.",
  "steps": [
    { "name": "ingestion-workflow", "type": "linear_workflow" },
    { "name": "analysis-agent", "type": "agent" },
    { "name": "distribution-workflow", "type": "custom_workflow" }
  ]
}