Variables and Context

This guide explains how to connect steps in a flow by passing data from one step to the next using variables.

How variables work

In a flow, each step can produce outputs that subsequent steps can use as inputs. This is how you chain steps together to create useful workflows.

For example:

1. An agent generates a chart and returns the image URL
2. An email tool sends that URL as an attachment

The email tool doesn’t know about the chart until you connect the agent’s output to the email’s input.

Types of inputs

When configuring a step’s inputs, you have two options:

Static values

A fixed value that stays the same every time the flow runs.

Use static values for:

• Email addresses that don’t change
• Configuration settings
• Default text

Variable references

A value that comes from a previous step’s output or a runtime parameter.

Use variables for:

• Passing data between steps
• Using runtime parameters from the Start step
• Dynamic content that changes each run

Selecting variables

When you click on an input field, a dropdown shows all available variables:

Variables are grouped by their source:

• Start — Runtime parameters defined in the Start step
• Step names — Outputs from previous steps (e.g., “Chart Generation Demo Agent”)

Click a variable to select it. The input field shows it’s using a reference:

Variable scope

Variables follow a simple rule: a step can only access outputs from steps that came before it.

Consider this flow:

[Start] → [Query Tool] → [Analyzer Agent] → [Slack Tool]

• Start parameters: accessible by Query Tool, Analyzer Agent, and Slack Tool
• Query Tool output: accessible by Analyzer Agent and Slack Tool only
• Analyzer Agent output: accessible by Slack Tool only
• Slack Tool output: not accessible by any other step (it’s last)

Caution

You cannot reference a step that hasn’t run yet. If you need data from a later step, you’ll need to reorder your flow.

Example: Chaining an agent to an email tool

Here’s a complete example of connecting variables between steps.

Step 1: Configure the agent

The agent receives the chart-prompt parameter from Start:

The agent produces several outputs:

• body — Email body content
• subject — Email subject line
• body_format — Format type (HTML/plain)
• attachment_asset_ids — Chart image IDs

Step 2: Configure the email tool

The email tool maps each input to the agent’s outputs:

After configuration:

Email input	Source
to	Static value: “user@example.com”
body	Chart Generation Demo Agent.body
subject	Chart Generation Demo Agent.subject
body_format	Chart Generation Demo Agent.body_format
attachment_asset_ids	Chart Generation Demo Agent.attachment_asset_ids

Result

When the flow runs, the email tool receives all the data it needs from the agent:

Runtime parameters

Runtime parameters are defined in the Start step and can be provided when the flow runs:

These parameters are available to all steps in the flow. They’re useful for:

• Making flows reusable with different inputs
• Allowing scheduled runs to use different values
• Testing flows with various scenarios

When you run a flow manually, you’re prompted to provide values:

Tips

Use descriptive step names. Clear names make it easier to identify which variable you need when configuring inputs.

Check available outputs. The dropdown shows all outputs a step produces. If you don’t see what you expect, check the agent or tool’s documentation.

Test incrementally. After connecting variables, run the flow to verify data passes correctly before adding more steps.

Next steps

• Scheduling Flows — Automate flow execution with schedules
• Flows overview — Return to the main flows introduction