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

# Personalization

> Give your Digital Twin its name, picture, instructions and look, and choose the AI model for each job, in the Personalization and AI Models tab.

The **Personalization and AI Models** tab of the admin editor is where you give your Digital Twin its identity — name, picture, backgrounds, About text and instructions — and choose which AI model does each of its jobs.

![Opening the Personalization and AI Models tab from Admin, Instances, Edit](https://mintcdn.com/praxisai/60S-TkaqdLQNeHho/images/admin-guide/admin-personalization-main.gif?s=3a43bc1843c4c383d8cc7040b21d7222)

**1. Open Admin**

Open the **Admin** section.

**2. Open Instances**

Select **Instances** in the admin menu.

**3. Edit your Digital Twin**

Select **Edit** on the Digital Twin you want to change.

**4. Open Personalization and AI Models**

Select the **Personalization and AI Models** tab. Save when you are done.

Most of these settings are also in **Instance Settings** (the chat's **+** menu, **Settings**, then your Digital Twin's name), where each change saves at once. See [Instance Settings](https://docs.praxis-ai.com/mdx/user-guide/profile-settings/instance-settings.md).

***

## Personalization (Digital Twin)

![The Personalization section of the admin editor](https://mintcdn.com/praxisai/60S-TkaqdLQNeHho/images/admin-guide/admin-personalizations-main-ui.webp?fit=max\&auto=format\&n=60S-TkaqdLQNeHho\&q=85\&s=af7e10c47223bc444554674c2cdd8474)

### Name & Avatar

> [accordion] Digital Twin Name (default Pria)
>
> The name people see in conversations, the gallery and across the interface. Left blank, it is **Pria**. Pick a name that reflects the twin's persona or expertise ("Professor Adams", "Campus Guide").

> [accordion] Picture URL
>
> The twin's avatar. Click the picture itself to upload a new one — there is no separate "Upload" button. Hover the picture for the camera overlay; the small **×** that appears on hover removes it. The picker crops a square image to a circle, so start from a square image at least 256 × 256 px with the subject centred. Left blank, the Pria logo is used.

> [accordion] Animated Picture URL (GIF)
>
> An animated avatar shown during conversations. Left blank, the static picture is used.

> [accordion] Background URL (Large Picture)
>
> The image behind the conversation area in light mode. Click the preview thumbnail to upload a new one (from a file or your webcam). The picker crops to widescreen `16:9`, so start from a landscape image at least `1920 px` wide.

> [accordion] Dark Mode Background URL (Large Picture)
>
> The same, for dark mode, so both themes look intentional. Left blank, the default dark background is used.

### Content & Behavior

> [accordion] About this Digital Twin
>
> A description of the twin's purpose, expertise and abilities, shown in the gallery. The label shows its length in characters.
>
> Select  **Generate** to have Pria write it from the twin's instructions and its active assistants. Update those first, then generate, for the most accurate result.

> [accordion] Use About for Welcome Screen
>
> Shows the About text and the twin's picture on the welcome screen instead of the default welcome content.

> [accordion] Persona mode
>
> The twin always speaks as its character: questions about what it can do are answered in its own voice, without platform guides or tool lists. A twin given instructions for the first time starts with Persona mode on.

> [accordion] Digital Twin Instructions
>
> The instructions that shape the twin's personality, expertise and behaviour — the most important setting for how it talks to people. The label shows their length in characters.
>
> The [Onboarding Questions](https://docs.praxis-ai.com/mdx/admin-guide/onboarding-questions.md) interview can write a first version for you. For methods and templates, see [Crafting Digital Twin Instructions](https://docs.praxis-ai.com/mdx/admin-guide/digital-twin-instructions.md).

> [accordion] Enable guardrail instructions
>
> Adds a separate block of rules, **Guardrail Instructions**, right after the instructions in every conversation, typed or spoken — for what the twin must always or never do. **Load default template** gives you a starting point when the box is empty.

### UI & Appearance

> [accordion] Guest UI (Minimum screen)
>
> A minimal, chat-only interface without the profile menu, sidebar or Digital Twin switching. Use it for public twins embedded in a website.

> [accordion] Disable Clipboard for Users
>
> Removes the copy and paste actions on answers for standard users — useful when the twin handles confidential content. Admins keep them.

> [accordion] Disable Audio Notes for Users
>
> Hides the microphone that lets standard users record spoken notes into their personal IP Vault. Admins keep it. Audio Notes is also hidden whenever **Guest UI** or **Disable File Upload for Users** is on. See [Audio Notes](https://docs.praxis-ai.com/mdx/user-guide/files-documents/audio-notes.md).
>
> > [!NOTE]
> >
> > If your Digital Twin existed before Audio Notes was introduced, this box was ticked for it so the new microphone wouldn't appear without warning. Untick it to offer Audio Notes. Newer Digital Twins have it on already.

> [accordion] UI CSS
>
> Custom CSS for your Digital Twin's interface — fonts, colours, layout. The **View template...** link next to the label opens a starter template with every available class. For example:
>
> ```css theme={null}
> :root {
>   font-size: 14px;
> }
> body {
>   font-size: 1rem;
> }
> ```
>
> See [UI Customization](https://docs.praxis-ai.com/mdx/admin-guide/ui-customization.md) for the full template reference.

> [accordion] UI Theme
>
> **-- User Directed --** (each person picks), **Dark Mode** or **Light Mode** for everyone.

A branding key, which applies custom branding to the Digital Twin, is set by the Praxis AI team: contact [humans@praxis-ai.com](mailto:humans@praxis-ai.com).

***

## AI Models

Each dropdown lists the models of the platform catalog suited to that job. **-- Default --** follows the platform default. See [AI Models](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md) for which model answers when several are set, deprecated models and reasoning effort.

### Custom Models

Lists the Digital Twin's active custom AI models. You add them in **Admin › AI Models** (see [Bring Your Own AI Model](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md#bring-your-own-ai-model-byom)). While a custom model is active for a job, that job's dropdown below is hidden.

### Conversation & Reasoning

| Setting | What it does |
| - | - |
| **Conversation** | The main model: it writes the answers, uses the tools and drives the twin. |
| **Fallback conversation model** | Answers when the conversation model fails before any reply started. Unset = the platform default. See [Fallback conversation model](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md#fallback-conversation-model). |
| **Conversation Completion Max Tokens** | The most tokens one answer may use, thinking included. **Unspecified (LLM Default)** lets the provider decide (often a low cap that turns thinking off); **Auto** lets Pria pick the model's own maximum; or choose a number. |
| **Reasoning Effort** | How much the model thinks before answering: **None (Disable thinking)**, **Low**, **Medium**, **High**, **Max**. Models that cannot think ignore it. See [AI Models](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md). |
| **Documents Summarization** | Summarizes uploaded documents and, when **Compact History** is on, older parts of long conversations. |
| **Extended Context (1M tokens)** | Turns on the 1 million token context window for the Claude models that offer it (Claude Sonnet 5 and Sonnet 4.6). The standard window is 200K tokens. |

> [!WARNING]
>
> **Conversation Completion Max Tokens:** Claude models need 4,000 or more to think before they answer. Models reached through Amazon Bedrock reserve about five times this number against a shared usage limit, so a very high setting can make requests fail with "Too many tokens" when many people chat at once. If that happens, lower it or set it to **Unspecified**.

> [!WARNING]
>
> **Extended Context** costs more per token from Anthropic. Turn it on only when the twin must read very large documents or keep very long conversations.

### Convo (Live Voice & Avatar Calls)

Shown when **Enable Convo** is on (see [Configuration](https://docs.praxis-ai.com/mdx/admin-guide/configuration.md)). The full provider guide is [Realtime Voice & Avatars](https://docs.praxis-ai.com/mdx/admin-guide/realtime-voice.md).

**Speech-to-Speech Model** decides the call experience. Anam models show a video avatar; the others are voice only:

| Provider | Models |
| - | - |
| **OpenAI GPT-Realtime** | `gpt-realtime-2.1 (Reasoning)` (platform default), `gpt-realtime-2.1-mini`, `gpt-realtime-2 (Reasoning)`, `gpt-realtime-1.5`, `gpt-realtime`, `gpt-realtime-mini` |
| **OpenAI GPT-Live** | `GPT-Live-1 (full-duplex)` (listens and speaks at the same time) |
| **Google Gemini Live** | `Gemini 3.8 Live`, `Gemini 3.1 Flash Live`, `Gemini 2.5 Flash Native Audio` |
| **xAI** | `grok-3-fast (voice)` |
| **ElevenLabs** | ElevenLabs Conversational AI — needs an Agent ID and API Key (a warning appears while they are missing; set them in the **Configuration and Integrations** tab, with the **ElevenLabs Connection Method**) |
| **Anam** | Anam Avatar |
| **Cartesia** | Cartesia Voice |

The fields below the model depend on the provider:

| Setting | Shown for | Default and choices |
| - | - | - |
| **Anam Response Model** | Anam | The model that writes the avatar's spoken answers. Empty = the conversation model. |
| **Voice Response Model** | Cartesia, GPT-Live | The same, for these providers. |
| **Voice Reasoning Effort** | All | **Default (None — thinking off for voice)**. Thinking time is silence in a call, so raise it with care. |
| **Voice** | OpenAI GPT-Realtime | **-- User's Choice --** (each person picks) or one of 10 voices: Cedar, Marin, Alloy, Ash, Ballad, Coral, Echo, Sage, Shimmer, Verse. |
| **Gemini Voice** | Gemini Live | **-- User's Choice --** or one of 30 voices, each labelled with its character (Puck — Upbeat, Kore — Firm…). |
| **xAI Voice** | xAI | **-- User's Choice --**, Eve, Ara, Rex, Sal or Leo. |
| **GPT-Live Voice** | GPT-Live | **-- User's Choice --** or a GPT-Live voice. |
| **Voice Activity Detection Eagerness** | OpenAI GPT-Realtime | How fast a pause ends the person's turn: **Low** (waits longer — good for people who pause mid-thought), **Medium (Default)**, **High** (fast back-and-forth). |
| **Noise Reduction** | OpenAI GPT-Realtime | **Off (Default)**, **Near Field** (headset or phone), **Far Field** (laptop or room microphone). |
| **Transcription Language (ISO 639-1)** | OpenAI GPT-Realtime | A two-letter code such as `en` or `es`. Blank = detect automatically. |
| **Interruption Sensitivity** | Gemini Live | **Low (Default)** ignores background noise but is harder to talk over — right for laptops in busy rooms. **High** stops the moment someone speaks — right for one-to-one headset use. It does not change how long Pria waits before deciding someone has finished. |
| **Speech Recognition Language (Deepgram Flux)** | Anam, Cartesia (also meeting bots and the public avatar embed) | **English only** (default) or **Multilingual** (de, en, es, fr, hi, it, ja, nl, pt, ru); multilingual also answers in the language the person spoke. |

Avatar look and greeting settings for Anam (placeholder image, introductory message and more) are in the **Configuration and Integrations** tab; see [Realtime Voice & Avatars](https://docs.praxis-ai.com/mdx/admin-guide/realtime-voice.md).

### Subagents

Pins the model, and the effort, for each subagent role. Leave them on the platform default unless you have a reason to change one. The switch that makes subagents available, and their budgets, are in the **Configuration and Integrations** tab. See [Configuration](https://docs.praxis-ai.com/mdx/admin-guide/configuration.md#subagent-role-models).

### Media

| Setting | What it does |
| - | - |
| **Image Analysis** | Reads images people upload: describes them and answers questions about screenshots, diagrams and photos. |
| **Image Generation** | Creates images from a description (for example GPT Image 2.5, a Gemini image model or Stability). |
| **Audio Generation** | Creates audio from a description. |
| **Video Generation** | Creates short videos (for example Amazon Nova Reel 1.1 or OpenAI Sora 2). |
| **Audio Transcription** | Turns speech in uploaded audio and video files into searchable text. |
| **Video Analysis** | Describes what a video shows on screen (Gemini models only). See [Video Analysis Model](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md#video-analysis-model). |
| **Text-to-Speech** | Reads answers aloud. |

### System

| Setting | What it does |
| - | - |
| **Embeddings Generation** | Turns file passages into the fingerprints used for search. Only `text-embedding-3-small` is offered. See [Knowledge & RAG Configuration](https://docs.praxis-ai.com/mdx/admin-guide/knowledge-rag-config.md#embedding-model). |
| **Moderation** | Checks messages when **Enable Moderation** is on. See [Content Moderation](https://docs.praxis-ai.com/mdx/admin-guide/moderation.md). |

***

## Related

* [Crafting Digital Twin Instructions](https://docs.praxis-ai.com/mdx/admin-guide/digital-twin-instructions.md) — methods and templates for the instructions
* [AI Models](https://docs.praxis-ai.com/mdx/admin-guide/ai-models.md) — model choice, reasoning effort, custom models
* [Realtime Voice & Avatars](https://docs.praxis-ai.com/mdx/admin-guide/realtime-voice.md) — voice providers and avatar setup
* [Knowledge & RAG Configuration](https://docs.praxis-ai.com/mdx/admin-guide/knowledge-rag-config.md) — file search and the knowledge graph
* [UI Customization](https://docs.praxis-ai.com/mdx/admin-guide/ui-customization.md) — the CSS template reference
* [Onboarding Questions](https://docs.praxis-ai.com/mdx/admin-guide/onboarding-questions.md) — the interview that drafts the instructions
* [Configuration](https://docs.praxis-ai.com/mdx/admin-guide/configuration.md) — the other Digital Twin settings


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.