> ## Documentation Index
> Fetch the complete documentation index at: https://docs.anyreach.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Condition step
> Branch the workflow based on data or AI judgment.
The Condition step splits the workflow into multiple branches. Execution routes down the first branch whose expression matches — or down the default branch if none match.
## How it works
The Condition step evaluates each branch's expression in order against the workflow context (`ctx`). The first expression that evaluates to true determines which branch executes next. If no expression matches, execution follows the **default condition** branch.
To add a new condition branch, hover over a condition node. A "+" icon will appear, click it to create an additional branch for your condition logic.
## When to use it
| Use Condition when | Don't use it when |
| -------------------------------------------------------------- | ------------------------------------------------------ |
| Routing based on a data value (status code, field value, tier) | The logic is computational — use a Code step |
| Classifying free-text with AI judgment | You already have structured data — use JSONata, not AI |
| Implementing a switch over an enum | |
## Inspector
Clicking on a condition branch in the workflow opens an editor in the inspector. Here, you can set a custom name for the branch and define its associated condition expression.
## Expression Types
Each branch in a Condition step uses an expression to determine when that branch should be followed. There are two supported types:
**JSONata expressions** — These are evaluated at runtime against the current workflow context (`ctx`). Use JSONata when you want to perform checks on structured data.
Examples:
```
{{ n1.name = "Adam Zampa" }}
{{ n1.score > 50 }}
```
You can also use the AI-powered expression generator in the editor to help you craft the right condition.
**AI expressions** — These let you phrase your condition as a natural language question to the LLM (Large Language Model). The model returns a true or false answer, allowing for more flexible or fuzzy logic that can use free text fields.
Example:
```
${ does {{n1.name}} contain the word "Adam"? respond yes/no }
```
Use JSONata for strict, structured logic where possible. Use AI expressions for complex, human-like reasoning needs.
## Branch ordering
Conditions are evaluated top to bottom. **First match wins** — once a branch matches, the others are not evaluated.
Order from most specific to least specific:
```
Branch 1: {{ n2.status_code = 200 }} ← most specific
Branch 2: {{ n3.status_code >= 400 }} ← catches all errors
default: ← fallback
```
## Default branch
Every Condition step has a default branch that executes when no other branch matches. Always wire it to a meaningful step — leaving the default unconnected means the workflow silently stops if nothing matches.