> ## Documentation Index
> Fetch the complete documentation index at: https://docs.eachlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Parameter References

> Wire outputs between steps using template variables.

## Template Syntax

Use `{{double braces}}` to reference workflow inputs and previous step outputs in your step parameters.

## Reference Types

| Reference             | Description              | Example                    |
| --------------------- | ------------------------ | -------------------------- |
| `{{inputs.field}}`    | Workflow input field     | `{{inputs.prompt}}`        |
| `{{step_id.output}}`  | Full output of a step    | `{{generate_text.output}}` |
| `{{step_id.primary}}` | Primary output of a step | `{{step1.primary}}`        |

## Examples

### Referencing Inputs

```json theme={"dark"}
{
  "step_id": "generate_text",
  "type": "model",
  "model": "openai-chatgpt-5",
  "params": {
    "system_prompt": "You are a helpful assistant",
    "user_prompt": "{{inputs.prompt}}"
  }
}
```

### Chaining Steps

```json theme={"dark"}
{
  "steps": [
    {
      "step_id": "step1",
      "type": "model",
      "model": "openai-chatgpt-5",
      "params": {
        "user_prompt": "{{inputs.prompt}}"
      }
    },
    {
      "step_id": "step2",
      "type": "model",
      "model": "flux-1-1-pro",
      "params": {
        "prompt": "{{step1.output}}"
      }
    },
    {
      "step_id": "step3",
      "type": "model",
      "model": "topaz-upscale-image",
      "params": {
        "image": "{{step2.primary}}",
        "scale": 2
      }
    }
  ]
}
```

### Embedding in Strings

Template variables can be embedded within larger strings:

```json theme={"dark"}
{
  "prompt": "A portrait of {{inputs.name}} in {{inputs.style}} style"
}
```

## Referencing Parallel Branch Outputs

When a parallel step completes, you can reference outputs from its branches using the branch step IDs:

```json theme={"dark"}
{
  "steps": [
    {
      "id": "step1",
      "type": "parallel",
      "branches": [
        {
          "name": "Video",
          "steps": [{"id": "step1_branch_0", "type": "model", "model": "veo-3", "params": {"prompt": "{{inputs.prompt}}"}}]
        },
        {
          "name": "Image",
          "steps": [{"id": "step1_branch_1", "type": "model", "model": "flux-2-max", "params": {"prompt": "{{inputs.prompt}}"}}]
        }
      ]
    },
    {
      "id": "step2",
      "type": "model",
      "model": "topaz-upscale-image",
      "params": {
        "image_url": "{{step1_branch_1.primary}}"
      }
    }
  ]
}
```

## Condition Expressions

Choice steps use `$` reference syntax (not `{{template}}` syntax) to evaluate conditions:

| Reference                | Description                   | Example                   |
| ------------------------ | ----------------------------- | ------------------------- |
| `$.inputs.field`         | Workflow input                | `$.inputs.email`          |
| `$.step_id.primary`      | Primary output of a step      | `$.step1.primary`         |
| `$.step_id.output`       | Full output of a step         | `$.step1.output`          |
| `$.step_id.output.field` | Nested field from step output | `$.step1.output.filename` |

```json theme={"dark"}
{
  "condition": {
    "expression": "$.step1.primary",
    "operator": "string_matches",
    "value": ".mp4"
  }
}
```

You can also access nested output fields:

```json theme={"dark"}
{
  "condition": {
    "expression": "$.step1.output.filename",
    "operator": "string_matches",
    "value": ".pdf"
  }
}
```

<Note>
  Condition expressions use `$.step_id.field` syntax, while step parameters use `{{step_id.field}}` template syntax. These are different reference systems.
</Note>

For the full list of condition operators (comparison, string, array, existence, and logical), check out [Workflow Structure: Condition Operators](/workflows/concepts/workflow-structure#condition-operators).

## Resolution

Template variables get resolved at execution time. Here's what the workflow engine does under the hood:

1. Evaluates steps in order
2. Executes parallel branches concurrently
3. Evaluates choice conditions and routes to the matching branch
4. Resolves template variables using actual values from inputs and completed steps
5. Passes the resolved parameters to the model

If a referenced step hasn't completed or doesn't exist, the execution will fail. So make sure your references point to steps that actually run before they're needed!