> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Text Agent Settings
> Step 1 of the wizard. Name your agent, pick the AI model, write the instructions that tell it how to behave, and dial in how creative its answers are.
This is **Step 1** for Text and Hybrid agents. Voice-only agents skip it and start with Voice Agent Settings. Once you save Step 1, the rest of the wizard unlocks and you can jump between steps freely.
## What this page looks like
Four cards stacked in the main column: **Identity**, **Model**, **Core Instructions**, **Creativity**. On the right (when your screen is wide enough), a live preview of the **Wittify chat widget** updates as you edit. On smaller screens, an **Agent Preview** button next to the page title opens the preview as a full-screen overlay.
## Identity card
The first card on the page. It holds one required field.
### Agent name *(Required)*
Also called **agent display name**. This is what you'll see in your dashboard. It's never shown to your customers.
| Tip | Example |
| ---------------------------- | ------------------------------------------------------------- |
| Pick something descriptive | `Website Support`, `Leads Bot · EN/AR`, `Sales Outreach , Q1` |
| Keep it under 100 characters | |
If you leave the name empty and click away, the field shows a red border and the message **Agent name is required**. The error stays until you type something.
A small hint sits under the field: *Choose a clear, descriptive name to identify this agent in your dashboard.*
## Model card
Pick the AI model that powers your agent. Six models are pinned as quick-pick tiles, and a **Browse all models** dropdown opens the full catalog of 27 models from 5 providers.
### Pinned models
A grid of model tiles. Each tile shows:
* The provider logo (OpenAI, Anthropic, Google, xAI, or Open Source)
* The model display name
* The provider name (small grey text under the model name)
* A type badge: **Instant** or **Reasoning**
The tile you've selected gets a colored border, a primary-tinted background, and a **Selected** badge in the corner. While the list is loading, you'll see six placeholder tiles pulsing in place.
**Instant** models are tuned for speed: they answer immediately. **Reasoning** models think through the problem before answering, which is slower but better for complex multi-step requests.
### Browse all models
Click the dropdown to see every model, grouped by provider:
| Provider | Includes |
| --------------- | ------------------------------------------- |
| **OpenAI** | GPT-5.4, GPT-4o, o4-mini |
| **Anthropic** | Claude Sonnet 4 (and other Claude versions) |
| **Google** | Gemini 2.5 Flash, Gemini 2.5 Pro |
| **xAI** | Grok variants |
| **Open Source** | Llama variants and others |
All models are served through Microsoft Azure (OpenAI, xAI, open source) and Google Cloud (Gemini, Anthropic Claude via Vertex AI). You'll see this in the **Browse all models** panel as the provider logos.
### What the meta chips mean
Below the tiles, three colored chips describe the model you've picked:
* **Context** , also called **context window**. How much the model can keep in mind in a single conversation. Bigger is better for long chats.
* **Arabic support** , how well the model handles Arabic. The chip color tells you at a glance: **green** = native or strong, **blue** = partial, **purple** = limited.
* **Latency** , how fast the model replies. **Green** = under 0.5 seconds, **blue** = under 2 seconds, **purple** = over 2 seconds.
### Advanced Settings (collapsible)
Click the **Advanced Settings** chip on the end edge of the meta row to open this panel. The chevron next to the chip rotates 180° when the panel is open.
| Setting | What it does | Default |
| ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- | ------- |
| **Max Tokens** , also called **output token limit**. The longest a single response can be (a token is roughly 0.75 of an English word). Range 256 to 32,000. | Raise it if your agent's answers are getting cut off. | 1,000 |
| **Top-P** , also called **nucleus sampling**. How varied the model's word choices are. Lower = more predictable, higher = more varied. Range 0 to 1. | Lower for policy bots, higher for creative writing. | 0.9 |
| **Reasoning Level** , how deeply the model thinks before answering. Three options: **Low**, **Medium**, **High**. Higher means more thinking time and a more thorough answer. | Only available for **Reasoning** models. Disabled (greyed out) with the label **Not available for instant models** when an Instant model is selected. | Medium |
Each setting has a **?** help icon next to its label. Hovering it shows a tooltip with a longer explanation.
**Most agents work great at the defaults.** Touch these only if you have a clear reason: answers cut off → raise Max Tokens. Want more deterministic phrasing → lower Top-P. Want deeper reasoning on complex requests → bump Reasoning Level.
## Core Instructions card *(Required)*
The brain of your agent. This is also called the **system prompt**. Think of it like onboarding a new employee, you're telling the agent exactly how to do its job. The editor inside this card is the **Prompt Editor**.
### What to write
Cover the basics:
* **Role and goal** , what the agent is for (support, sales, bookings, lead qualification).
* **Tone and style** , friendly, professional, concise, formal.
* **Language and dialect** , be specific. Say *English (US)* or *Arabic , Saudi* or *Arabic , Egyptian*. Vague language instructions are the #1 cause of dialect drift.
* **What to do, what NOT to do** , when to escalate, what topics to refuse, how to redirect off-topic questions.
* **Knowledge sources** , point to the FAQs, files, or URLs the agent should rely on. You'll wire the actual content in the Knowledge Base step.
You have up to 2,500 characters. A live counter sits at the bottom-end edge of the editor.
### Helpful buttons in this card
* **Enhance with AI** , a one-click button that rewrites your draft into a stronger version using AI. After it runs, two new buttons appear:
* **Keep** , uses the new version.
* **Revert** , goes back to your original.
* **Quick template chips** , one-click starters: **Customer support**, **HR assistant**, **Government services**. The active template is highlighted in the primary brand color.
* **More templates** , a dropdown that groups templates by category (sales, ops, internal, government, telecom, etc.).
If you leave Instructions empty and click away, the field shows a red border and the message **Instructions are required**.
## Creativity card
A slider that controls **temperature**, the AI term for how creative the agent is when phrasing answers. Range is 0 to 1. The slider on screen runs from 0 to 10 (each step is 0.1).
| Slider value | Label | Use for |
| ------------ | --------------------- | ------------------------------------------------------------------------------- |
| 0.0 to 0.3 | **Precise** (green) | Policy bots, documentation, factual answers, anything where consistency matters |
| 0.4 to 0.6 | (middle) | Balanced. Good for general assistants. |
| 0.7 to 1.0 | **Creative** (purple) | Marketing copy, brainstorming, storytelling |
The slider track and the number readout change color to match the band you're in. A **?** help icon next to the **Creativity** title shows a tooltip explaining the trade-off, and a hint below the slider updates with each value.
For support, ops, and policy bots, keep this at **0.5 or lower**. For marketing or creative writing, push above **0.7**.
## The bottom action bar
Every wizard page has the same bottom bar, called the **wizard bottom nav**:
| Button | What it does |
| ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| **Cancel** (before your first save) | Exits the wizard. No agent is created. |
| **Exit** (after your first save) | Leaves the wizard. If you have unsaved changes, an unsaved-changes dialog asks before leaving. |
| **Back** | Goes to the previous step. Disabled on Step 1. |
| **Step dots** | Click any unlocked step to jump to it. Locked steps appear muted (you can only unlock them by saving Step 1 first). |
| **Save draft** | Saves what you have without moving forward. A **Draft saved** toast confirms. |
| **Next , *Voice Agent Settings*** (or whatever the next step is called) | Saves the current step and moves to the next one. The button shows the *name* of the destination step. |
If you try to close the browser tab with unsaved changes, the page asks if you really want to leave.
## The wizard sidebar
The left sidebar is called the **wizard sidebar**. It shows:
* The list of steps for this agent type (6 for Text, 6 for Voice, 7 for Hybrid). Each step has an icon, title, and a status dot: empty before save, **green check** after save.
* An **Agent name** card after your first save. The card updates only when you save, not as you type.
* The shared plan card, platform links, and help menu at the bottom (same as the dashboard sidebar).
You can collapse the sidebar from full width (240px) to a slim icon-only strip (60px) using the toggle button below the logo.
On mobile (under `lg`), the sidebar becomes a slide-in drawer from the start edge. Open it with the hamburger menu in the topbar.
## The wizard topbar
The topbar shows:
* **Breadcrumbs** , Wittify logo / agent type / current step name.
* **See Tutorial** button , see below.
* **Theme toggle** , switches between light and dark.
* **Language switcher** , flips the wizard between English and Arabic, preserving your place.
* **Profile dropdown** , Settings, User Activities, Plans & Subscription, Log out.
* **Notifications bell** (no badge during wizard).
## Step loading
When you switch steps, the main column briefly shows three pulsing placeholder cards instead of the form. This happens while the wizard fetches the destination step's saved state. Once the data arrives, the form fills in automatically.
## The wizard tutorial
Click the **See Tutorial** button in the topbar to open a guided spotlight tour of the page you're on. The tutorial highlights each card in turn (Identity → Model → Core Instructions → Creativity) with a tooltip explaining what to do.
You can close the tutorial in three ways:
* **Finish** , finishes the tour.
* **Skip** , dismisses for this session.
* Click the backdrop or press **Esc** , dismisses for this session.
Choosing **Don't show again** stops it from auto-opening on future first visits, but the **See Tutorial** button still re-opens it whenever you want.
## How saving works
After your first save, your work is preserved if you refresh the page. Every step is saved on its own when you press **Save draft** or **Next**. You can come back days later and pick up where you left off.
The wizard never persists secrets like API keys or webhooks to your local browser, only to your account. So even on a shared computer, draft credentials stay safe.
## Variant differences
After saving Text Agent Settings, the next step is **Voice Agent Settings**. Hybrid is the only type that has both. The Hybrid wizard has 7 steps total.
This is Step 1. The next step is **Appearance**. Voice settings, voice deployment, and call telephony are entirely hidden for text-only agents. The Text wizard has 6 steps total.
This step **does not exist** for voice-only agents. They start with Voice Agent Settings, which has its own Agent Name field. The Voice wizard has 6 steps total.
## Common questions
Yes. Open your agent's [Agent Settings](/en/Systems/AI-Agents/Manage-Agent/Agent-Settings) → **Text Settings** and pick a different model. Conversations already in flight finish on the old model.
The wizard saves your progress as you go. After your first save, refreshing reloads your work where you left off.
The model you picked is an **Instant** model. Switch to a **Reasoning** model (like o4-mini or Claude Sonnet 4) and the pills will activate.
Yes. The text box automatically handles Arabic and English. The model behaves best when you state the **language and dialect** explicitly inside the prompt itself.
The right side of the page shows a live preview of the **Wittify chat widget**. On smaller screens, click **Agent Preview** next to the page title to open the preview as a full-screen overlay.
Yes, while you're moving between steps. Use **Save draft** any time you want to keep your work without moving forward.
## Next
Pick a voice and tune call behavior. Hybrid agents only.
Brand the chat widget, set per-language text, configure the launcher and welcome banner.