A complete starter project for building voice AI apps with LiveKit Agents for Python.

The starter project includes:

• A simple voice AI assistant based on the Voice AI quickstart
• Voice AI pipeline based on OpenAI, Cartesia, and Deepgram
  • Easily integrate your preferred LLM, STT, and TTS instead, or swap to a realtime model like the OpenAI Realtime API
• Eval suite based on the LiveKit Agents testing & evaluation framework
• LiveKit Turn Detector for contextually-aware speaker detection, with multilingual support
• LiveKit Cloud enhanced noise cancellation
• Integrated metrics and logging

This starter app is compatible with any custom web/mobile frontend or SIP-based telephony.

Clone the repository and install dependencies to a virtual environment:

cd agent-starter-python
uv sync

Set up the environment by copying .env.example to .env.local and filling in the required values:

• LIVEKIT_URL: Use LiveKit Cloud or run your own
• LIVEKIT_API_KEY
• LIVEKIT_API_SECRET
• OPENAI_API_KEY: Get a key or use your preferred LLM provider
• DEEPGRAM_API_KEY: Get a key or use your preferred STT provider
• CARTESIA_API_KEY: Get a key or use your preferred TTS provider

If you want to use Google Cloud Text-to-Speech instead of Cartesia, you'll need to set up Google Cloud credentials:

1. Create a service account in your Google Cloud Console
2. Enable the Text-to-Speech API for your project
3. Grant the service account the necessary IAM roles:
   • Go to IAM & Admin > IAM in the Google Cloud Console
   • Find your service account and click the edit (pencil) icon
   • Add the following role: Cloud Text-to-Speech API User (roles/texttospeech.user)
   • Alternatively, you can use the more permissive Editor role, but the specific TTS role is recommended for security
4. Download the service account JSON credentials file
5. Convert the credentials to base64 format and add to your .env.local:

# Convert your Google Cloud credentials JSON to base64
python3 -c "
import json
import base64

# Read your downloaded credentials file
with open('path/to/your/credentials.json', 'r') as f:
    creds = json.load(f)

# Convert to base64
creds_json = json.dumps(creds)
creds_b64 = base64.b64encode(creds_json.encode()).decode()
print('Add this line to your .env.local:')
print(f'GOOGLE_APPLICATION_CREDENTIALS_B64={creds_b64}')
"

Then add the base64-encoded credentials to your .env.local:

GOOGLE_APPLICATION_CREDENTIALS_B64=your_base64_encoded_credentials_here

Note: The base64 encoding method is recommended over raw JSON in environment variables because it avoids issues with special characters and line breaks in the private key.

You can load the LiveKit environment automatically using the LiveKit CLI:

lk app env -w .env.local

Before your first run, you must download certain models such as Silero VAD and the LiveKit turn detector:

uv run python src/agent.py download-files

Next, run this command to speak to your agent directly in your terminal:

uv run python src/agent.py console

To run the agent for use with a frontend or telephony, use the dev command:

uv run python src/agent.py dev

In production, use the start command:

uv run python src/agent.py start

Get started quickly with our pre-built frontend starter apps, or add telephony support:

For advanced customization, see the complete frontend guide.

This project includes a complete suite of evals, based on the LiveKit Agents testing & evaluation framework. To run them, use pytest.

uv run pytest

Once you've started your own project based on this repo, you should:

1. Check in your uv.lock: This file is currently untracked for the template, but you should commit it to your repository for reproducible builds and proper configuration management. (The same applies to livekit.toml, if you run your agents in LiveKit Cloud)

2. Remove the git tracking test: Delete the "Check files not tracked in git" step from .github/workflows/tests.yml since you'll now want this file to be tracked. These are just there for development purposes in the template repo itself.

3. Add your own repository secrets: You must add secrets for OPENAI_API_KEY or your other LLM provider so that the tests can run in CI.

This project is production-ready and includes a working Dockerfile. To deploy it to LiveKit Cloud or another environment, see the deploying to production guide.

This project is licensed under the MIT License - see the LICENSE file for details.