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

# AI

> Send an instruction mid-workflow and wait for the model response. Exact configuration, ports, placement, and execution behavior.

**Node type:** `ai_prompt` · **Version:** `1` · **Category:** Data

Send an instruction mid-workflow and wait for the model response. This node does not declare an external side effect.

## Configuration

| Field          | Type   | Default                                                                      | What it controls                                                                                   |
| -------------- | ------ | ---------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
| `connectionId` | string | `""`                                                                         | Reference to an AI connection whose credential is stored in the operating-system credential vault. |
| `prompt`       | string | `""`                                                                         | Instruction sent to the selected model. It can be mapped from an earlier node output.              |
| `systemPrompt` | string | `"You are a helpful workflow assistant. Return a concise, useful response."` | System instruction that defines the model's role for this workflow step.                           |
| `temperature`  | number | `0.2`                                                                        | Creativity value from 0 to 1.                                                                      |
| `maxTokens`    | number | `1200`                                                                       | Maximum number of output tokens requested from the model, from 64 to 32000.                        |
| `timeoutMs`    | number | `90000`                                                                      | Maximum time for this operation in milliseconds.                                                   |

<Note>The workflow pauses at this node until the selected provider responds, the request is cancelled, or the configured timeout is reached.</Note>

<Note>The credential remains in the host vault. The prompt, system instruction, and mapped context are sent to the selected provider.</Note>

<Note>AI nodes run on the desktop local runner and return the response text, provider-reported usage, and model ID.</Note>

## Workflow JSON

The editor stores this node with the following implemented default configuration:

```json title="Default node configuration" theme={"system"}
{
  "type": "ai_prompt",
  "version": 1,
  "configuration": {
    "connectionId": "",
    "prompt": "",
    "systemPrompt": "You are a helpful workflow assistant. Return a concise, useful response.",
    "temperature": 0.2,
    "maxTokens": 1200,
    "timeoutMs": 90000
  }
}
```

Values may be entered literally or mapped from an earlier compatible output when the inspector exposes a mapping control. See [Variables and data mapping](/workflows/variables-and-data).

## Inputs

| Port     | Type     | Required | Description |
| -------- | -------- | -------- | ----------- |
| `prompt` | `string` | Yes      | Instruction |

## Outputs

| Port       | Type     | Required | Description |
| ---------- | -------- | -------- | ----------- |
| `response` | `string` | No       | AI response |
| `usage`    | `object` | No       | Token usage |
| `model`    | `string` | No       | Model       |

## Where it can run

* Desktop local runner

## Test and inspect

Use **Test node** in the editor to preview this step with the current configuration. A full run records resolved inputs, outputs, logs, duration, and any artifacts in the execution inspector. Side-effecting or destructive nodes can require an additional confirmation or approved workflow permission.