> For the complete documentation index, see [llms.txt](https://docs.borndigital.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/configuration-tab.md).

# Configuration tab

Input messages and attachments, AI provider and model, temperature, reasoning, history, response handling, streaming and timeout.

The **Configuration** tab holds the settings that rarely need to change. **Reset to default** restores the provider and model, timeout, history, response variable, response handling, temperature and input messages. It does not touch the Instructions, tools, response format or reasoning.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2F8QRhQHRr5gWEzeHSO7jP%2Fai-node-configuration.png?alt=media&amp;token=c49a1b64-12b9-4ab4-ad2b-4bc97b13ca68" alt="The Configuration tab"><figcaption><p>The Configuration tab</p></figcaption></figure>

## Modifying input for AI response

The messages the model answers to, sent after the Instructions and the history.

* The default is one **User** message with `{current_utterance}` — what the user just said.
* Click the role label to switch between **User** and **Assistant**. A chain of User/Assistant messages can show the model an example of the expected answer.
* Variables are written as `{variable_name}`.
* **Each iteration** (on by default): the message is sent every time the node runs. When it is off, the message is sent only the first time the conversation reaches this node.

{% hint style="warning" %}
The user's utterance is **not** added automatically. If you replace the default message, keep `{current_utterance}` in it, for example `Answer the customer's question: {current_utterance}`.
{% endhint %}

{% hint style="info" %}
A message with **Each iteration** off is remembered as already sent only when **Save in history** is on. With **Save in history** off, it is sent on every run.
{% endhint %}

### Attachments (files in user input)

Use the paperclip of a message to attach variables that contain files — images or documents the model should read.

* The variable contains an `https://` URL reachable from the internet, or a list of such URLs.
* If the file must not be exposed to the internet, store it where the platform can reach it (for example a blob storage) and name the variable with the prefix `binary_`, for example `binary_invoice`. The platform downloads the file and sends its content to the model; the content is neither logged nor stored.
* Supported: PNG, JPEG, WEBP, non-animated GIF, PDF, DOC, DOCX. Up to 50 MB per request.

## AI provider

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FNdWLvoSfuxV4BZGzJEwE%2Fai-node-ai-provider.png?alt=media&amp;token=f137608a-d12a-4d73-a6d7-969d37cddccf" alt="AI provider and model"><figcaption><p>AI provider and model</p></figcaption></figure>

* **Provider** and **Language Model** – the lists are maintained by the platform and differ between environments. Each model may show a tag, for example its speed.
* For an **on-premise** provider, fill in **Endpoint**, **API Key** and **Language Model**.
* The **Latency** indicator next to the model estimates how long it takes until the user hears the first sentence, including the model's first token and the speech settings of the project. It assumes **Stream sentences** is on for longer answers.

{% hint style="info" %}
Your organization may be limited to one approved provider. Choose another provider only after testing the answer quality in your language.
{% endhint %}

## Preparing the response

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FNr1fg7FksdFOH8HDkjxj%2Fai-node-preparing-response.png?alt=media&amp;token=93ee6c90-e6b8-41fe-b259-67b104583f8d" alt="Preparing the response with reasoning settings"><figcaption><p>Preparing the response with reasoning settings</p></figcaption></figure>

| Setting                              | Meaning                                                                                                                                                                        | Default                        |
| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------ |
| **Temperature**                      | 0–2. Higher values give more varied answers, lower values more consistent ones. Locked at 1 for GPT-5 and GPT-6 models.                                                        | 1                              |
| **Response format**                  | **Text** or **Json**. For Json, the Instructions must contain the word *json* and describe the expected structure.                                                             | Text                           |
| **Reasoning → Effort**               | Shown for reasoning models. How much the model thinks before answering. More effort means better answers to complex tasks, but higher latency and cost.                        | the lowest option of the model |
| **Reasoning → Summary**              | **Auto** or **Detailed**. A summary of the model's reasoning is stored in `reasoning_summary` of the response variable (not in streaming mode). It is never shown to the user. | Auto                           |
| **Use History** / **History length** | How many previous exchanges the model sees.                                                                                                                                    | on, 4                          |

The **Effort** options depend on the model: GPT-5 models offer *minimal, low, medium, high*; newer models offer *none* and *extra high*, some pro models start at *medium*. The editor shows only the options the selected model accepts.

For voicebots, keep the effort low: every step up adds latency before the first word.

### History

* **History length** counts the user's messages: with 4, the model sees the last 4 user messages and everything that followed them (answers, tool calls).
* **0** – no history; every turn is answered on its own.
* **Unlimited** – the whole conversation is kept at the provider. Works only with **Save in history** on.
* History is **shared by all AI nodes** of the conversation. A second AI node with history on also sees what the first one discussed — useful for specialist agents, but keep it in mind when one node only classifies or extracts.

## Handling the response

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2F4aIniyVhDtAqk4esJvbF%2Fai-node-handling-response.png?alt=media&amp;token=7a1736f9-d2be-49f3-a9b7-a93b8d2fae6e" alt="Handling the response with streaming and a first response message"><figcaption><p>Handling the response with streaming and a first response message</p></figcaption></figure>

| Setting                   | Meaning                                                                                                                                                                                                                              | Default        |
| ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------------- |
| **Save in variable**      | Name of the response variable. See [Response variable and routing](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/response-variable-and-routing.md).                                            | `gpt_response` |
| **Save in history**       | Adds this turn to the AI conversation history. Switch off for background nodes (classification, extraction) whose exchange should not influence later answers.                                                                       | on             |
| **Send response to user** | Sends the answer to the user. Switch off when the node works in the background and another node speaks.                                                                                                                              | on             |
| **Stream sentences**      | Sends the answer sentence by sentence while the model is still writing — the user hears the first sentence much sooner. Works in channels that support streaming (voice, the new chat bubble); elsewhere the answer is sent at once. | off            |

### First response messages

With **Stream sentences** on, you can add **First response messages**, for example *"Let me check that for you."* If the model has not produced any text within **First streamed sentence delay in ms** (100–5,000 ms, default 500), one of these messages is played at random, once per turn. Use it for nodes that call tools, where the first sentence can take longer. The message becomes part of the answer text.

## Timeout

How long the node waits for each request — every call to the model and every call to an MCP server. Choose 5, 10, 30, 60 or 120 seconds, or **Custom**. Default: 30 s.

If the model does not answer in time, the response text is empty and the response variable contains `error`. Plan a path for this in the flow — see [Response variable and routing](https://docs.borndigital.ai/digital-agent/conversation-flow/nodes-explained/ai-node/response-variable-and-routing.md).
