> For the complete documentation index, see [llms.txt](https://docs.sentifyd.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.sentifyd.io/features/markdown.md). # Training Avatars ## Training Avatars ### Overview In Sentifyd, our training is highly adaptable. You can assign and reassign training to any avatar without dependence on a specific avatar. Start with our pre-made templates or design your own training program from scratch. ### Creating a Training Creating training for your avatar involves three main steps, each of which is managed through different tabs in the training form. #### **1. Persona Tab** In this tab, you'll define the persona of your avatar. This step is crucial as it sets the tone and character of the avatar's responses. * **Training Name**: Give your training a meaningful name for easy identification. * **Response Length**: Restrict the response length by specifying a maximum word count. * **Style**: Choose from various styles such as creative, formal, friendly, or funny, depending on the desired personality of the avatar. * **Persona Role**: Define the role of the avatar (e.g., Customer Support, Virtual Salesperson). * **Persona Description**: Provide a detailed description of the persona to guide the AI in generating appropriate responses. * **Conversation Examples:** Provide up to 25 Q\&A examples the avatar should always answer from directly. These are injected into the system prompt, so they ground the avatar's behavior without needing a knowledge-base lookup — ideal for persona details, canned responses, preferred phrasings, and short factual answers. {% hint style="info" %} **CSV Upload Guidelines**: When uploading a CSV, make sure it adheres to the specified format, with distinct columns for user inputs and AI responses. {% endhint %} #### **2. Knowledge Sources Tab** This tab is where you provide sources such as documents or a website address to enable the chatbot to respond based on your sources. The sources ground the chatbot and enable it to provide specific answers based on your needs. * **Documents:** You may upload up to five documents in the following formats: PDF, Word, Excel, PowerPoint, Text, or HTML. Each document can be a maximum of 16 MB. * **Website URL:** Enter a publicly accessible website URL. The website content is extracted and indexed for the avatar. You can re-index the URL at any time to refresh the content. **How website indexing works:** * Starting from the URL you provide, Sentifyd follows links within the same site up to **3 levels deep** and indexes up to roughly **150 pages**, using your site's sitemap when available to pick the most relevant pages. * Only publicly accessible pages are indexed — pages behind logins or paywalls are not. * Indexing is a **snapshot**: when your website content changes, re-index the URL from the training page to refresh the avatar's knowledge. * For large sites, point the avatar at the most relevant section (e.g., `https://example.com/help`) rather than the homepage, so the page budget is spent on content that answers your users' questions. #### **3. Tools Tab** The **Tools Tab** allows you to extend your AI agent’s capabilities by enabling additional tools it can use during live conversations. Tools enhance your agent’s functionality—allowing it to perform searches, send data, and even interface with your own systems in real time. **✅ Always-Enabled Tools** These tools are built-in and cannot be disabled, as they form the foundation of Sentifyd’s interactive experience: * **Control 3D Avatar**\ Allows the agent to animate and control the 3D avatar—managing gestures, facial expressions, posture, and avatar view (e.g., full-body or head-only). * **Control Sentifyd Widget**\ Enables the agent to manipulate elements of the chat interface—such as toggling captions, switching between voice and text modes, and ending or restarting conversations. *** **🌐 Web Search Tool** Allow your agent to safely access real-time web information. This tool enables the avatar to perform controlled web searches, expanding its ability to answer questions beyond your internal data sources. *** **📩 Send Support Request Tool** When the agent can’t resolve a user’s issue, this tool enables it to escalate gracefully. It will: * Prompt the user for contact details (email and phone) * Automatically send a support request to your predefined support email **Setup required**: You must provide a support email address in the configuration. *** **⚙️ MCP Tools – Custom Backend Integrations** MCP (Model Context Protocol) tools allow your agent to securely interact with your own APIs, systems, or services. Perfect for advanced workflows, these tools enable your agent to: * Send and receive data via HTTP or SSE connections * Execute real-time API calls (e.g., to booking systems, internal dashboards, or CRMs) * Stream live updates into the chat or avatar response **Configuration options:** * **Connection Type**: Streamable HTTP or Server-Sent Events (SSE) * **Endpoint URL**: Your service URL * **Headers (JSON)**: Optionally, add custom request headers (e.g., `{"Authorization": "Bearer TOKEN"}`) {% hint style="info" %} You don't need to build anything to use MCP tools — no-code platforms like Zapier can host them for you. See [Adding MCP Tools to Your Avatar](/features/adding-mcp-tools-to-your-avatar.md) for setup options, from no-code to custom servers. {% endhint %} #### Submitting the Training Fill out the forms and review your inputs to ensure accuracy. When ready, submit your training for processing. It may take several minutes to fully index and process. Check the training page to monitor the indexing status.