Building an AI Digital Sales Assistant
Author： Alec Lee | AX Global Strategist & Full-Stack Explorer. If you find any areas for improvement in this article, please feel free to contact me.
We’ll walk you through building an AI-powered digital sales assistant capable of:
🧠 Listening
🗣️ Speaking
🤖 Thinking
🎭 Performing
All through a real-world use case: Luxury Car AI Sales Avatar – Vendy
This project is built using the following products: Dify × TEN × Trulience × Azure × Agora × OpenAI Integration Guide
Introduction
What are the six platforms that power a “Listen–Think–Speak–Perform” AI Avatar?
This project is centered on the TEN platform, integrating six key systems to create Vendy, a luxury car AI sales avatar with full conversational capability.
Vendy can:
- 👂 Listen to questions
- 🧠 Think and respond
- 🗣️ Speak answers
- 🧍 Present herself as a digital human
👇 Below is the architecture diagram showing the full system integration:
🧩 Platform Roles Overview
👇 This diagram outlines the role of each platform, their integration modules, and the collaboration workflow:
🧩 Roles of Each Platform in the System
👇 The diagram below illustrates how each platform contributes within the system — including their integration modules and collaborative workflow:
📌 Platform Functions Overview
- Dify + OpenAI
Builds the intelligent agent core. It connects to both internal enterprise data and the OpenAI API, providing a unified conversation API interface to TEN.
- Trulience
Enables digital human rendering. It provides a digital human ID and access token to TEN for visual avatar presentation.
- Agora
Handles voice and video communication. It provides App ID and Certificate to TEN for real-time interaction.
- Azure Speech
Offers speech-to-text (STT) and text-to-speech (TTS) services. It provides the following parameters to TEN:
-
AZURE_STT_KEY -
AZURE_STT_REGION -
AZURE_TTS_KEY -
AZURE_TTS_REGION -
TEN (Central Platform)
Serves as the orchestration hub, integrating and managing APIs, keys, and tokens from all platforms.
TEN enables Vendy, the AI Sales Avatar, to conduct intelligent conversations and real-time interactions.
📌 Note: Each of the platforms above can be substituted with alternative open-source or SaaS services depending on your use case.
The current combination is a reference setup, designed for learning and reusability.
Building the Central Brain of the AI Sales Agent
We’ll start with three simple steps in Dify to build the “core intelligence” of our AI Sales Avatar — based on internal company data.
Our demo scenario: a luxury car sales assistant named Vendy.
🧩 Step 1: Register a Dify Account & Configure Model API
👉 Go to the official Dify website, Click “Get Started” in the top right corner to begin the registration process.
You can sign up or log in using one of the following methods:：GitHub、Gmail，or Hotmail.
After logging into Dify, follow these steps:
Click “Settings > Model Provider” in the left-hand menu.
On the page, locate the OpenAI module and click Install to set it up.
In the OpenAI module, click the blue link labeled “Get your API Key from OpenAI”. You’ll be redirected to the official OpenAI platform：https://platform.openai.com/api-keys。
On the OpenAI platform, click “+ Create new secret key” to generate a new API key. Copy the generated key.Return to Dify, and paste the key into the OpenAI module under the following fields:
🧩 Step 2: Upload Internal Documents for the AI Sales Avatar
Once the model configuration is complete, your environment is ready. You can now begin building the knowledge base for your AI Sales Avatar in Dify.
- In the Dify console, click “Knowledge” from the top menu.
- Then click “Create Knowledge” to create a new knowledge dataset.
On the following page, choose your preferred method for uploading content.
We recommend using the “Import from file” option for the most straightforward setup.
👇 As shown in the illustration below:
Please upload your prepared internal documents — such as product brochures, sales manuals, FAQ sheets, and more.
If you haven’t prepared relevant materials yet, you can also use ChatGPT to help generate content.
👇 For example, we generated a Word document using ChatGPT that introduces several luxury car brands (e.g., Lexus, Audi, BMW, etc.):
After downloading the document, return to the Dify upload page,
select “Import from file”, and upload the prepared file.
✅ Once the upload is successful, click the “Next” button in the bottom-right corner to proceed to the next configuration step.
The default settings are suitable for most use cases.
We recommend keeping them as they are and simply clicking “Save & Process” to complete the setup.
⚙️ (Optional) If you later wish to improve the AI assistant’s comprehension accuracy,
you can adjust parameters based on your specific business needs.
✅ If you see the following page, it means your document has been successfully uploaded and parsed by the Dify system.
At this point, we’ve completed the knowledge base setup — your AI Sales Avatar now has its foundational memory in place.
🧩 Step 3: Create an AI Application and Connect the Knowledge Base
- In the Dify console, click the top menu item “Studio”.
- Then, click the “Create” button in the upper-right corner to start building a new App.
In the pop-up window, fill in the following fields:
- App Name & Icon：Example:
AI Sales Avatar - Description： Your name is Vendy, the AI Clerk of Luxury Cars. Your task is to provide information about luxury cars to the customers based on the knowledge.
👇 The settings interface looks like this:
After clicking “Create”, you’ll be taken to the App’s Orchestrate editing interface.
In the INSTRUCTION section, enter content consistent with the earlier Description, such as:
Your name is Vendy, the AI Clerk of Luxury Cars. Your task is to provide information about luxury cars to the customers based on the knowledge.
Next, scroll down to the Knowledge section.
Select the knowledge document we uploaded earlier as the reference material for the AI application.
Once the configuration is complete, use the input box on the right side of the Orchestrate interface to test your AI agent. If the responses reference the content from your uploaded document, it means the knowledge base integration is working correctly.
After confirming everything works as expected, click the “Publish” button in the upper-right corner to officially launch your AI application.
After publishing the application, click “API Access” from the left-hand sidebar menu.
Then click the “Create Secret Key” button in the upper-right corner.
This API Key is essential for integrating your AI application with the TEN platform in the next step.
Please make sure to copy and store it securely.
👇 The secret key generation interface looks like this:
🎉 Congratulations!
You’ve now completed the setup of your AI Sales Avatar’s core intelligence and successfully obtained the API Key required for integration.
Next, we’ll move into the Trulience platform, where we will design the digital appearance and performance capabilities of your AI Sales Avatar.
In other words — we will give her a face, expression, and presence, enabling her to not only “listen, speak, and think,” but also to appear on screen and interact with users in a human-like manner!
📚 To learn more about Dify, visit the official documentation:
👉 https://docs.dify.ai/en/introduction
Create the Digital AI Avatar
🧩 Step 4: Register a Trulience Account and Get Your Digital Avatar ID & Token
Now it’s time to give your AI Sales Avatar a visual identity!
We’ll use the Trulience platform to generate a digital human and retrieve the necessary Avatar ID and Access Token for integration.
✅ Follow these steps:
1️⃣ Go to the Trulience website: https://trulience.com
Click the “Start For Free” button on the homepage to begin registration.
👇 Sample interface:
2️⃣ After registration, you’ll be redirected to the dashboard.
From the left-hand menu, click “IMAGE AVATARS”.
Choose a female character to represent Vendy’s appearance,
then click the green “Create Avatar” button to generate the digital human.
👇 Sample interface for avatar selection and creation:
3️⃣ When you see the “Avatar created” message, it means your digital human has been successfully created.
👇 Confirmation screen after successful creation:
3️⃣ Rename the default avatar name (e.g., Amanda) to Vendy using the field on the left panel.
Once renamed, you’ll see her unique Avatar ID displayed below.
👇 Example view:
4️⃣ Next, click the “Connect” button at the bottom of the screen.
Your browser will redirect to a new page.
From the URL, copy the string that comes after token= —
this is your Access Token.
🎉 Congratulations!
You’ve successfully completed the visual appearance setup for your AI Sales Avatar,
and obtained both the Avatar ID and Access Token.
Enabling Voice Capabilities for Digital AI Avatar
🧩 Step 5: Prepare Three Key Configurations Before Running TEN Agent Locally
Before integrating the Dify agent and Trulience digital human into the TEN platform,
we need to set up the accounts and credentials required for voice communication and speech processing.
Specifically, you’ll need to prepare the following three components:
✅ 1. Agora Real-Time Voice Communication Parameters
TEN uses Agora to enable real-time voice interactions (i.e., “she can speak”).
You’ll need the following parameters:
AGORA_APP_ID=（App ID）AGORA_CERTIFICATE=（Certificate）
👉 How to get it:
Log in to the Agora Console.
Create a new project, and you’ll be able to access the required parameters.
👇 Sample view of the Agora Console:
After logging in, click “Projects” from the left sidebar.
Then, click “Create New” in the top-right corner to create a new project.
👇 Project creation interface:
📝 Input Field Guide:
- Project Name: Use something like
AISalesAvatar - Use Case: Select an appropriate option, e.g.,
Enterprise Collaboration - Authentication Mode: Choose Secure Mode: APP ID + Token (Recommended)
Once completed, click “Submit” to create the project.
👇 After successful creation, you will be able to access your App ID and Certificate:
Once the project is created, you’ll be redirected to the project management page.
In the project list, click the copy icon next to App ID to copy it.
👇 Example interface:
After copying the App ID, click the ✏️ pencil icon (Edit) on the right side of the project row to enter the Project Details page.
Scroll to the bottom of the page to find the Primary Certificate section.
Click the copy icon to copy the project’s Certificate.
👇 Example view:
At this point, you’ve successfully obtained the Agora App ID and Primary Certificate,
completing the required setup for enabling real-time voice communication in the TEN platform.
📚 For more information, refer to the official Agora documentation:
✅ 2. Azure Speech Parameters (Speech-to-Text + Text-to-Speech)
Azure handles two critical voice functions:
- STT (Speech-to-Text): Converts the user’s spoken words into text
- TTS (Text-to-Speech): Converts the AI’s text responses into speech
You’ll need to prepare the following four environment variables:
AZURE_STT_KEYAZURE_STT_REGIONAZURE_TTS_KEYAZURE_TTS_REGION
Next, fill in the required fields under the Basics tab:
👇 Reference guide for the “Basics” section:
For most cases, selecting the default option “Public access” is sufficient,
which allows the service to be accessed over the internet.
🧩 Optional Step:
You can add resource tags — for example, project=ai-agent — to help with future organization and management.
👇 Example view of the Tags and Networking configuration screen:
📝 After reviewing and confirming that all information is correct,
click the “Create” button to start deploying the Azure OpenAI resource.
👇 Deployment confirmation screen:
👇 Resource deployment in progress:
After deployment is complete, navigate to Azure AI Foundry.
You’ll see the following welcome screen, which allows you to quickly create a project and select a model:
👇 Azure AI Foundry welcome interface:
Go to the “Model Catalog” page and select Azure AI Speech.
👇 Model selection interface:
After clicking the Azure-AI-Speech model, you’ll be taken to its details page.
Click the “Use Service” button at the top of the page to jump to the Speech Playground.
👇 Speech service launch interface:
Inside the Speech Playground, select your speech service resource (as shown below).
You’ll then be able to view the corresponding API Key and Region.
👇 Speech service resource selection and key display:
Click the “Text to Speech” tab to enter the Voice Gallery,
where you can preview various voice styles and select the one that best fits your AI Avatar.
👇 Voice Gallery interface for selecting speech styles:
Click the “View Code” button to copy the current speech service’s API Key and Region.
👇 Example interface for copying credentials:
🎉 Congratulations!
You’ve now completed all the required parameter preparations. For running the TEN platform and connecting your AI Sales Avatar.
🧱 What’s Next: Launching the TEN Agent Platform Locally
Now we’re ready to launch the TEN Agent platform on your local machine and connect all the components we’ve prepared:
- 🧠 Intelligent Agent (via Dify)
- 🔊 Voice Capabilities (via Azure Speech + Agora)
- 👩💻 Digital Avatar (via Trulience)
Together, these form a truly real-time, embodied AI Sales Assistant
that can listen, think, speak, and perform — end to end.
Let’s build it!
📚 For more details on deploying Azure Speech and OpenAI services, refer to the official documentation:
Debugging the Voice Interaction System for Digital Humans
🧩 Step 6: Set Up the TEN Agent Development Environment on macOS
To run and test the AI Sales Avatar’s real-time voice interaction system, we first need to set up the TEN Agent development environment on your Mac.
Prerequisites
🔧 Required Tools & Installation Guide
| Tool | Purpose | Installation Method |
|---|---|---|
| Docker | Container environment for running TEN | Download from the official Docker site |
| Git | Clone and manage the TEN codebase | brew install git |
| Node.js | Run the front-end (Playground) | brew install node |
✅ ① Install Docker
💡 Docker is the core runtime environment for the TEN platform — all modules will run in containerized form.
Follow these steps:
-
Open the official Docker website: 👉 https://www.docker.com/products/docker-desktop
-
Download the
.dmginstaller for macOS and complete the installation. -
After installation, launch Docker from your Applications folder.
-
Wait until the 🐳 Docker icon in the top menu bar turns green, indicating it’s running successfully.
-
Open your Terminal and run the following command to verify the installation:
Verify the installation: docker -v. If you see a version number, Docker is installed correctly.
✅ ② Install Git
💡 Git is a version control tool used to clone and manage the TEN framework codebase.
- Install Git using Homebrew: brew install git. Verify the installation: git -v. If a version number appears, Git is installed successfully.
✅ ③ Install Node.js
💡 Node.js is required to run the TEN Playground frontend locally.
- Install Node.js using Homebrew: brew install node Verify the installation: node -v. If a version number appears, Node.js has been installed successfully.
✅ ④ Clone the TEN Agent Codebase
Once Docker, Git, and Node.js are installed, the next step is to clone the TEN Agent source code and prepare the .env environment configuration.
🚀 Step 1: Clone the TEN Agent GitHub repository
Open your Terminal and run: git clone https://github.com/TEN-Framework/TEN-Agent.git
This will create a local folder named TEN-Agent, containing all the necessary code and configuration files for the platform.
👇 Directory structure example:
📦 Step 2: Navigate to the project directory and copy the environment variable template
Change into the ai_agents subdirectory and copy the example .env file:
cd TEN-Agent/ai_agents cp .env.example .env
The .env file contains the environment variables required to run the TEN Agent.
Next, you’ll need to manually edit this file and input the necessary credentials, including:
- Azure STT/TTS Keys and Regions
- Agora App ID & Certificate
👇 Example view of the .env configuration file:
✍ Step 3: Edit the .env File and Fill in the Following Variables
Use a text editor (we recommend Cursor or VS Code) to open the .env file.
Scroll to the end of the file and add the following variables — replacing the values with your own credentials:
Agora App ID and Certificate AGORA_APP_ID=your_agora_app_id AGORA_APP_CERTIFICATE=your_agora_certificate
Azure STT Parameters AZURE_STT_KEY=your_stt_key AZURE_STT_REGION=your_stt_region
Azure TTS Parameters AZURE_TTS_KEY=your_tts_key AZURE_TTS_REGION=your_tts_region
👇 Example view of the .env file being edited:
✅ Step ⑤: Start the TEN Agent Container and Run the Server
Once your .env file has been properly configured, the next step is to launch the TEN Agent container environment.
🐳 ① Start the Container
Navigate to the ai_agents directory and start the Docker container services:
cd ai_agents docker compose up -d
This command will launch the following container services:
ten_agent_dev: Main development container for running the Agent
ten_agent_demo: Example agent instance
ten_agent_playground: Frontend interface for interactive testing
👇 Project structure example:
🔧 ② Enter the Container and Compile the Agent Server
Run the following command to access the development container:docker exec -it ten_agent_dev bash
🔧 ③ Run the Initialization Script to Register the Agent.
Once inside the container, execute the following command to register the Agent:task use
👇 Example of the project structure after initialization:
This command will automatically read and link the Agent’s configuration files
(such as manifest.json and property.json),
and install all necessary dependencies.
👇 Example output of the execution process:
✅ After completing task run, open your browser and go to the following address to launch the frontend:
If everything is running correctly, you should see the interface below:
- Click on the “Graph” tab at the top of the page.
- From the dropdown menu, select your desired Agent instance.
👇 Example UI view:
➡️ voice_assistant_intergrated_stt (Auto Start)，
Then, click the “Save” button in the bottom-right corner to save your configuration.
👇 Sample interface view:
✅ Replace the Text-to-Speech (TTS) Module with Azure Engine
- Locate the TTS module in the graph (e.g.,
tts_default) - Right-click on the module and select “Replace Node with”
- From the dropdown list, choose
azure_tts
This completes the replacement of the TTS module with the Azure-based engine.
👇 Example view of replacing the TTS module with Azure TTS:
🎙️ Customizing the Digital Human’s Voice (TTS Settings)
You can customize the voice of your digital avatar to match her personality and appearance.
Microsoft provides a full list of supported voice names in its speech service documentation.
👉 Visit the following link to browse available voices:
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=tts#multilingual-voices
In this example, we selected an English female voice:
🗣️ en-US-AshleyNeural
In the Azure TTS module, set the following field:
Azure_synthesis_voice_name=en-US-AshleyNeural
🔁 Replace the LLM Module with Dify API to Connect Your Intelligent Agent
- Right-click on the
llmmodule and select “Replace Node with” - From the popup list, choose
dify_python
This will replace the default large language model (LLM) in TEN Agent with the Dify-powered intelligent agent interface.
👇 Example view of the replacement process:
2️⃣ After replacing the module, right-click the newly created dify_python node again
and select “Update Properties” to open the configuration panel.
- In the API Key field, enter the API Key you created on the Dify platform
- In the Greeting field, you can optionally customize a welcome message (e.g., a voice greeting)
👇 Example view of the Dify module property settings:
✅ Start TEN Agent and Connect Trulience AI Avatar
🚀 Launch the Frontend Interface
Once configuration is complete, open your browser and navigate to the following address
to access the TEN Agent launch interface:
👇 Example view of the launch interface:
🧠 Select the Agent Graph (Workflow)
-
Click “Select Graph” in the upper-right corner, and from the dropdown menu choose:
➤voice_assistant_integrated_stt -
Check the box for “Enable Trulience Avatar” and fill in the following fields:
- Avatar ID (obtained during Trulience avatar creation)
- Avatar Token (copied from the URL after clicking “Connect” in Trulience)
-
After completing the fields, click “Save changes” to apply the settings.
👇 Example view of the configuration screen:
🗣️ Start Voice Interaction and Verify Digital Human Connectivity
✅ Click the “Connect” button in the upper-right corner to run the final test.
If everything is configured correctly, the following should happen:
-
🧍 The digital human avatar will load and appear on screen
-
🗣️ She will greet you with natural speech and ask a question, such as:
“Which luxury cars can I introduce to you today?” -
🤖 Based on her response, you can verify whether she successfully retrieved and understood content from the Dify knowledge base.
👇 Example view of the running interaction:
🎯 Congratulations!
You now have a fully functional AI Sales Avatar!
✅ A conversational brain powered by OpenAI / Dify
✅ Natural voice synthesis (female voice) via Azure
✅ A digital human appearance rendered by Trulience
✅ Real-time voice interaction built on the TEN platform
🌟 This is the complete and essential chain of an AI Agent:
Listen → Think → Speak → Perform → Connect
📚 To learn more, refer to the official TEN documentation:
https://github.com/TEN-Framework/TEN-Agent
👉 https://theten.ai/docs/ten_agent/architecture_flow
🛠️ Next Step: Expansion Suggestions (Based on AX Architecture)
Building upon your current AI Sales Avatar,
you can follow the AX Architecture roadmap to further enhance its capabilities in real-world business scenarios.
The goal is to gradually integrate a full-cycle loop of:
Data → Model → Agent → User Interface → Full Business Lifecycle Management
🧑💼 Embed on the Company Homepage as a Virtual Brand Representative
By integrating Trulience’s D-Human and Avatar capabilities,
you can deploy your digital human onto the company homepage or web application.
Through natural voice-based Q&A, the avatar can guide visitors, capture their intent,
and serve as a user-centric entry point that connects to:
Marketing InfoCustomer Lifecycle
📦 Integrate with Databases & Order Management Systems to Evolve into an “AI Sales Representative”
Connect the Avatar with internal enterprise systems such as:
- CRM, ERP, OA, and order management tools (e.g., Salesforce, SAP, Notion DB, etc.)
This enables the AI agent to perform end-to-end tasks like:
- Sales lead management
- Inventory checks
- Contract generation
Serving business processes such as:
Contract LifecycleAsset Lifecycle
🔗 Integrate with Core Enterprise Platforms: From “Individual Use” to “Organizational Collaboration”
You can further modularize the Agent PF (platform) capabilities
and embed them into the Data Lakes and workflow management platforms of various BUs.
This unlocks cross-functional integration scenarios such as:
- AI + Finance
- AI + HR
- AI + R&D
- AI + SNS (Collaboration)
Ultimately, it paves the way for a company-wide AI transformation and upgrade.
✨ Final Reflections
I’ve always believed that AI should not merely be a cold, powerful tool —
but also a warm companion to humanity,
an extension and amplifier of our individual potential.
In the near future, we may see more digital companions like “Vendy” quietly emerge across the world —
beings who listen, think, respond, and express,
using the power of technology to uphold human dignity and creativity.
If you, too, wish to join this global journey of co-creation,
we warmly invite you to explore, experiment, and share —
so that every individual can grow alongside AI,
and together, we may step into the AX era of:
“One Against a Thousand — Leaving No One Behind.”
🌱 A Humble Beginning
This document is just a small experiment — a seed of possibility.
May it quietly take root, and gently begin to sprout
in a future where humans and AI coexist in harmony.