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

# Prompt

> Generate typed data with an AI prompt against upstream pipeline data.

<Note>This node is in **beta**. Its behavior may change in future releases.</Note>

The **Prompt** action node runs a free-form AI instruction over your pipeline data and returns structured, typed JSON that matches a schema you define. Use it when you need the model to reason, summarize, classify, or synthesize a value that a fixed extraction schema cannot express on its own.

## When to use Prompt

* You want to **derive a new value** from already-extracted data: a risk score, a one-line summary, a normalized category, a computed recommendation.
* You need **AI judgment** over the document or upstream output, not a literal field pulled from the page. Use [Extract](/nodes/extract) when the value is present on the document; use Prompt when it must be inferred.
* You want the result as **typed JSON** you can reference downstream, not free text. The output schema guarantees the shape.

## Configuration

| Field             | Type            | Required | Description                                                                                                                                                                                                                                                 |
| ----------------- | --------------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Instruction**   | template editor | Yes      | The prompt sent to the model. Supports template expressions, so you can feed in upstream values, e.g. `Summarize the risk of invoice {{extract.payload.invoiceNumber}} for a total of {{extract.payload.total}}`. The run fails if the instruction is empty |
| **Precision**     | select          | Yes      | The model tier used to run the instruction: `Small`, `Medium`, or `High`. Higher precision uses a stronger model and costs more credits. Default: `Small`                                                                                                   |
| **Output Schema** | schema editor   | Yes      | A JSON schema describing the shape of the result. The model is constrained to return data that matches it. The run fails if no output schema is set                                                                                                         |

The instruction and output schema are both required at run time. A Prompt node with either one blank fails validation and cannot be activated.

## Output

The result is typed JSON that matches your **Output Schema**, available to downstream nodes under the node's payload. Reference a field as `{{promptNodeName.payload.<field>}}`, where `promptNodeName` is the node's name in your pipeline.

For example, with an output schema defining `riskLevel` (string) and `reasons` (array of strings), a downstream node can read `{{prompt.payload.riskLevel}}` and `{{prompt.payload.reasons}}`.

## Inputs and outputs

**Allowed inputs:** action nodes only, including Extract, Parse, Classify, Route, Merge, Transform, Review, HTTP, Loop, Variable, Store, Reconcile, and another Prompt node. Connect a Prompt node after the step that produces the data you want the model to reason over.

**Output:** a payload object shaped by your output schema.

## Credits

Prompt nodes bill by **Precision**, a flat cost each time the node runs (not per page):

* **Small:** 1 credit
* **Medium:** 2 credits
* **High:** 6 credits

See [credits](/admin/credits) for the full per-node pricing breakdown.

## Related

<CardGroup cols={2}>
  <Card title="Extract action" icon="sparkles" href="/nodes/extract">
    Pull structured data that is present on the document
  </Card>

  <Card title="Transform action" icon="arrows-rotate" href="/nodes/transform">
    Reshape or script-process data without AI
  </Card>

  <Card title="Schema design" icon="sitemap" href="/guides/schema-design">
    Design the output schema that shapes the result
  </Card>

  <Card title="Expressions and filters" icon="braces" href="/guides/expressions">
    Template syntax for the instruction field
  </Card>
</CardGroup>