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

# Assistant node: LLM calls in your pipeline

> Run Claude, GPT, Gemini, and other LLMs as a step in your Knouds pipeline. Supports text generation, vision analysis, and structured list output.

The Assistant node runs a large language model as a step in your pipeline. You can use it to generate text from a prompt, analyze an image, rewrite or expand output from another node, or extract structured data to pass downstream.

## Available models

The model picker lists every enabled LLM model in your workspace. Three models are available by default:

* **Claude Sonnet** — Anthropic's flagship model with strong reasoning and vision
* **GPT-5.4 Mini** — OpenAI's compact model, fast and cost-effective for everyday tasks
* **Gemini 3.1 Flash Lite** — Google's lightweight Gemini variant, optimized for fast responses

Your workspace administrator can add more models — any OpenAI-compatible endpoint, additional Anthropic or Gemini variants, or custom hosted LLMs — through the Node Builder.

## Input sockets

| Socket      | Type          | Description                                                     |
| ----------- | ------------- | --------------------------------------------------------------- |
| `prompt-in` | Text (orange) | The user-turn prompt. Overrides what is typed in the node body. |
| `image-in`  | Image (pink)  | Image input for vision-capable models                           |
| `system-in` | Text (orange) | System prompt. Overrides what is set in the side panel.         |

When a socket is connected, it takes precedence over the value typed directly on the node. This lets upstream nodes — like a Text Box with static instructions, or an Image Generator — drive the assistant automatically.

## Output sockets

| Socket                      | Type          | Description                                                  |
| --------------------------- | ------------- | ------------------------------------------------------------ |
| `analysis-out` / `text-out` | Text (orange) | The model's text response                                    |
| `items-out`                 | Any (purple)  | Structured list output (when Export mode is set to **List**) |

## Node body

The node body has two views you can toggle between using the icons in the top-right corner of the card:

* **Prompt view** — shows what will be sent to the model. If the `prompt-in` socket is connected, this displays the incoming value.
* **Result view** — shows the model's response after a run. You can edit this text directly; edits persist and flow downstream.

While the model is running, the body shows an animated thinking indicator. Once the response starts arriving, it streams in character by character.

## Controls on the node

At the bottom of the node:

* **Model picker** — select the LLM to use.
* **Export mode** — choose **Export as Text** (returns the full response as a string) or **Export as List** (splits the response into individual items for use with a List node).
* **Settings** icon — opens the side panel.
* **Run** button — execute the node.

## Side panel inspector

Click the **Settings** icon to open the side panel. From there you can:

* Write or edit the **system prompt** in a full-width text area
* Configure the **prompt** for the current run
* Adjust model-specific parameters such as temperature, max tokens, and reasoning effort (availability depends on the model)

## Credit cost

The node header shows the estimated credit cost before you run (for example, `~30 cr`). The cost comes from the model definition and updates if an administrator changes the pricing.

## Common use cases

<CardGroup cols={2}>
  <Card title="Prompt generation" icon="pencil">
    Connect a Request Input → Assistant → Image Generator to let the LLM expand a short user description into a detailed image prompt before sending it to the generator.
  </Card>

  <Card title="Image analysis" icon="eye">
    Connect an Image Generator (or Media Upload) → Assistant to describe, classify, or extract information from images using a vision-capable model.
  </Card>

  <Card title="Text transformation" icon="repeat">
    Chain two Assistant nodes to translate, rewrite, or summarize the output of the first before passing it downstream.
  </Card>

  <Card title="Structured extraction" icon="list">
    Set Export mode to **List** and connect the `items-out` socket to a List node to turn a numbered response into individual items for batch processing.
  </Card>
</CardGroup>

<Tip>
  To analyze a generated image, connect the `image-out` socket of an Image Generator to the `image-in` socket of an Assistant node. Then connect the Assistant's `text-out` to a Response Output to return the analysis as text.
</Tip>