> For the complete documentation index, see [llms.txt](https://documentation.proto.cx/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://documentation.proto.cx/docs/modules/ai-agents/workflows-and-actions/llm-processing.md).

# LLM processing

`Action` — Sends a prompt to an LLM mid-chat and stores the response in a variable for downstream actions.

## What it does

Generates or transforms text at any point in the flow — summarising what a person said, classifying their request, or drafting a tailored reply — and makes the result available to the next action.

## When to use it

To process free-form input intelligently mid-chat, such as extracting structured fields from a long description or deciding which path to take based on sentiment.

## Settings

| Setting                | Description                                                                                                                                                                               |
| ---------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Input**              | The content the LLM acts on. Use plain text or a chat variable in curly brackets, e.g. `{description}`.                                                                                   |
| **Model**              | The LLM to use: DeepSeek V2, GPT-4o, Qwen 2.5, or Llama 3.2. Defaults to DeepSeek V2.                                                                                                     |
| **Temperature**        | Controls response randomness. Range 0.0–2.0, default 1. Lower values produce more deterministic output.                                                                                   |
| **System prompt**      | The standing instruction that shapes how the LLM responds, separate from the input. See [example system prompt](/docs/modules/ai-agents/knowledge/example-system-prompt.md) for guidance. |
| **Use knowledge info** | When toggled on, the LLM can also draw on the AI agent's [Knowledge](/docs/modules/ai-agents/knowledge.md) training sources.                                                              |
| **Resulting variable** | The chat variable that stores the LLM's response. Must start with a letter and contain only letters, numbers, or underscores.                                                             |

Press **Save** to confirm edits.

## Outcome branches

The action branches into **Success**, **Error**, and **Timeout**. Nest subsequent actions under each as needed.

## Example

```
LLM processing: "complaint details"
• Input:              {description}
• Model:              GPT-4o
• System prompt:      extraction instructions
• Resulting variable: complaint_details
└─ Success
   └─ Create ticket → Description: {complaint_details}
```

## Related

{% content-ref url="/pages/yRuISyntS8UJxaE5kgNX" %}
[Branch](/docs/modules/ai-agents/workflows-and-actions/branch.md)
{% endcontent-ref %}

{% content-ref url="/pages/rTdnuIG08MgrRXkV3Alv" %}
[Set variable](/docs/modules/ai-agents/workflows-and-actions/set-variable.md)
{% endcontent-ref %}