Gemini Live Avatar Agent Demo

This repository contains a complete, self-contained template for running a Gemini Live Agent with an Avatar Configuration Plugin and a beautiful, glassmorphic Web Client (avatar_client.html). It is built using Google's Agent Development Kit (ADK) and the Gemini Live API.

This demo enables real-time bidirectional audio and video streaming of a Gemini digital avatar (e.g., "Kai").

live-avatar-agent-demo/
├── README.md                  # This documentation file
├── .gitignore                 # Standard Python gitignore rules
├── requirements.txt           # Python dependency specifications
└── live_avatar_agent/         # The agent application package
    ├── __init__.py
    ├── agent.py               # The main ADK agent definitions
    └── avatar_client.html     # Real-time WebSocket/MediaSource web client

Before getting started, make sure you have the following installed and configured:

1. Python 3.11+
2. Google Cloud SDK (gcloud)
3. Google Cloud Project: You need an active Google Cloud Project with the Vertex AI API enabled.
4. Google Cloud Credentials: Authenticate your terminal using:

   gcloud auth application-default login

Follow these steps to set up and run the demo locally:

Clone or download this folder, navigate into it, and create a Python virtual environment:

# Create virtual environment
python3 -m venv .venv

# Activate virtual environment
# On macOS/Linux:
source .venv/bin/activate
# On Windows:
# .venv\Scripts\activate

Install the dependencies:

pip install -r requirements.txt

Launch the ADK web server pointing to the current directory:

# Set your active Google Cloud Project ID
export GCP_PROJECT="your-google-cloud-project-id"

# (Optional) Set your Vertex AI Search Engine/App ID if you want to test website search
export VERTEXAI_ENGINE_ID="your-search-engine-id"

# Run the ADK server
adk web .

You should see output similar to this:

+-----------------------------------------------------------------------------+
| ADK Web Server started                                                      |
|                                                                             |
| For local testing, access at http://127.0.0.1:8000.                         |
+-----------------------------------------------------------------------------+

Since the client communicates directly with localhost:8000 via WebSockets, you can open the HTML client directly in any browser:

• Double-click the file avatar_client.html or run:

  # On macOS:
  open live_avatar_agent/avatar_client.html
  # On Linux:
  xdg-open live_avatar_agent/avatar_client.html

Once opened, you will see the glassmorphic Gemini Avatar Client interface.

• It will automatically connect to the ADK backend server.
• Click the Microphone icon at the bottom of the page to start speaking, or type a text message in the chat input.
• The avatar will respond in real-time with bidirectional video and audio streaming!

The agent is configured to use the DiscoveryEngineSearchTool (Vertex AI Search) for answering internal queries.

• If you have an active website data store, configure VERTEXAI_ENGINE_ID inside agent.py or set it via environment variables before launching.
• If you don't need search, you can remove the DiscoveryEngineSearchTool import and tool from the tools=[...] list inside agent.py.

The agent uses the AvatarConfigPlugin to request the avatar modality.

• Currently, options include 'Kai' and 'Cora'.
• You can change avatar_name = 'Kai' in the before_run_callback method of the AvatarConfigPlugin class in agent.py.

To share this demo with other engineers or customers:

1. Initialize the local Git repository:

   git init
   git add .
   git commit -m "Initial commit for Live Avatar Agent Demo"

2. Create a new repository on GitHub (e.g., named live-avatar-agent-demo).

3. Push the local repository to GitHub:

   git remote add origin https://github.com/<your-github-username>/live-avatar-agent-demo.git
   git branch -M main
   git push -u origin main