> For the complete documentation index, see [llms.txt](https://docs.artific.nl/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.artific.nl/en/for-administrators/assistants/settings.md).

# Assistant settings

The main configuration screen for an assistant: who it is, how it writes, and what it is allowed to do. Reached through **AI assistants → Assistants → ⋮ → Settings**, or the **Settings** button on the chat screen.

The screen is grouped into sections. Everything on it is saved together with the **Save** button at the bottom.

## Personality

This section decides what the assistant is and how it behaves. It carries more weight than any other setting on the page.

| Setting                 | What it does                                                                 |
| ----------------------- | ---------------------------------------------------------------------------- |
| **Name**                | The assistant's display name, shown everywhere including the chat window.    |
| **Short description**   | One line describing what the assistant is for. Appears on the overview card. |
| **Instructions for AI** | The written brief that shapes every answer.                                  |

### Writing Instructions for AI

This is the single highest-leverage field in the platform. Treat it as an induction note for a new colleague who is competent but knows nothing about your organisation.

Cover, in plain language:

* **Role.** Who the assistant is and who it is helping.
* **Scope.** What it should answer, and what it should refuse or redirect.
* **Sources.** Whether it must answer only from the attached documents, or may use general knowledge as well. Be explicit. This is a common cause of disappointment.
* **Behaviour when unsure.** What to do when the answer is not in the documents. "Say you do not know and point the user to <hr@example.com>" is far better than leaving it open.
* **Format.** Short answers, bullet points, always cite the policy name, and so on.

Write in full sentences and be specific. Vague instructions produce vague assistants.

## Messages

| Setting              | What it does                                                                                                                                  |
| -------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- |
| **Opening message**  | The greeting the assistant sends when a conversation starts.                                                                                  |
| **Tone of voice**    | Applies one of your organisation's saved writing styles. See [Tone of voice](https://docs.artific.nl/en/for-administrators/tone-of-voice.md). |
| **Creativity**       | How much variation the assistant allows itself, on a five-step scale: **Minimum**, **Low**, **Medium**, **High**, **Maximum**.                |
| **Reasoning effort** | How much a model that reasons thinks before it answers. Only shown for models that support it.                                                |
| **Show reasoning**   | Shows users a **Reasoning** block above the answer. Only shown for models that support it.                                                    |

### Choosing a creativity level

| Level                  | Use it for                                                                                          |
| ---------------------- | --------------------------------------------------------------------------------------------------- |
| **Minimum** / **Low**  | Factual work: policy answers, support, anything where the same question should get the same answer. |
| **Medium**             | The general-purpose default.                                                                        |
| **High** / **Maximum** | Brainstorming, drafting marketing copy, generating options.                                         |

For a knowledge assistant, lower is almost always better. High creativity makes an assistant more likely to fill gaps with plausible-sounding invention.

Creativity applies to models that answer directly. A model that reasons before it answers does not use it, and the screen says so beside the setting.

## General

| Setting                                          | What it does                                                                                                           |
| ------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------- |
| **Theme**                                        | The visual style of the chat window. See [Themes](https://docs.artific.nl/en/for-administrators/assistants/themes.md). |
| **Model**                                        | Which AI model powers the assistant.                                                                                   |
| **Chat with an employee (Direct Chat add-on)**   | Lets a user ask to be handed over to a colleague mid-conversation.                                                     |
| **Allow file upload in chat messages**           | Gives users a paperclip button so they can attach documents to a question.                                             |
| **Allow image upload in chat messages**          | Lets users attach images. Requires a model that can read images, and requires file upload to be on as well.            |
| **Allow assistant to query tables in documents** | Lets the assistant read tabular data inside documents rather than only the prose.                                      |
| **Authentication needed**                        | Requires the person chatting to be signed in. Relevant mainly for assistants embedded on a website.                    |
| **Show references**                              | Adds a **Sources used** block under answers, showing which documents the answer came from.                             |
| **Assistant visible to users**                   | The on/off switch. While it is off, the assistant is greyed out on the overview and cannot be used.                    |
| **Enable Projects**                              | Lets users create projects with this assistant. See [Projects](https://docs.artific.nl/en/for-everyone/projects.md).   |
| **Allow users to set personal instructions**     | Lets each user add their own standing instructions for this assistant.                                                 |
| **Enable voice (beta)**                          | Lets users hold a spoken conversation instead of typing.                                                               |
| **Mask personal data**                           | Removes personal data from messages before the assistant processes them.                                               |

{% hint style="success" %}
Turn **Show references** on for any assistant that answers from documents. Being able to see which document an answer came from is what turns a plausible answer into a trustworthy one, and it is the fastest way for users to spot when something is out of date.
{% endhint %}

### Choosing a model

The **Model** field opens a picker listing the models available to your organisation. Each is tagged with what it can do:

| Tag            | Meaning                                                                  |
| -------------- | ------------------------------------------------------------------------ |
| **Fast**       | Optimised for speed. Good for high-volume, straightforward questions.    |
| **Multimodal** | Can interpret images as well as text. Required for image upload.         |
| **Tools**      | Can use tools and connectors, and is required for Direct Chat.           |
| **Memory**     | Can use the memory feature.                                              |
| **Reasoning**  | Thinks before it answers. With these models, **Creativity** is not used. |

If you pick a model that cannot do something you have switched on, the screen tells you so directly, with a **Select a different model** shortcut. Watch for:

* **Model does not support tools and direct chat**: your tools and handover will not work.
* **Model does not support memory**.
* **Model does not support the image search tool**: image search in the knowledge base will not function for this assistant.

{% hint style="info" %}
Which models appear in the picker depends on your organisation's plan and configuration. Your account manager can tell you what is available to you.
{% endhint %}

Models are added and retired over time. A retired model disappears from the picker, so check your assistants and Toolbox elements when that happens, and choose a successor. A different model does not always behave like its predecessor: test a switch first with [quality control](https://docs.artific.nl/en/for-administrators/quality.md) before you change the model of an assistant people rely on.

### Reasoning

Many models think before they answer. With such a model, two settings appear under **Messages**.

**Reasoning effort** sets how much the model thinks: **Off**, **Low**, **Medium** or **High**, or the default. More reasoning gives better answers to hard questions, but takes longer and costs more. For straightforward questions a low level is enough. Some models always reason; for those, reasoning cannot be switched off. While a model reasons, **Creativity** is not used.

**Show reasoning** adds a block above each answer showing how long the model thought and which steps it took, such as searching the knowledge base or using a tool. That makes an answer easier to follow.

{% hint style="warning" %}
Everyone who sees the answer also sees the reasoning, and the reasoning can paraphrase the assistant's instructions. Be careful with this setting on assistants that anyone can open, such as one on your website. Switching it off does not hide reasoning that is already saved.
{% endhint %}

### Settings that may cost extra

Switching on **Chat with an employee (Direct Chat add-on)** or **Enable voice (beta)** raises a confirmation: *Additional costs may apply*. Both are add-ons. Turning them off again never prompts. Contact your account manager before enabling them if you are not sure about the commercial side.

## Chat widgets

| Setting          | What it does                                                                                                                            |
| ---------------- | --------------------------------------------------------------------------------------------------------------------------------------- |
| **Email canvas** | Adds an e-mail composition panel inside the chat, so a user can turn an answer into a drafted message without leaving the conversation. |

## Saving

**Save** stores everything on the page. If a colleague changed the same assistant while you had the screen open, you get a **Settings have changed** dialog offering a link to the current version and an **Overwrite** button. Read the current version before overwriting. The other person's changes are lost otherwise.

## Deleting an assistant

At the bottom of the page, **Delete this assistant** removes it permanently.

{% hint style="danger" %}
Deletion cannot be undone, and any chat widget embedded on a website using this assistant stops working immediately. You must type the assistant's exact name to confirm.
{% endhint %}

## Next

* [Advanced settings](https://docs.artific.nl/en/for-administrators/assistants/advanced-settings.md): access restrictions, session limits, answer review and fallbacks.
* [Sub-agents](https://docs.artific.nl/en/for-administrators/assistants/sub-agents.md): letting one assistant delegate to another.
