A flexible voice AI agent system that seamlessly integrates PipeCat's real-time voice capabilities with LangChain's powerful AI orchestration and LangGraph's multi-agent routing.

This system provides a bridge between PipeCat's voice infrastructure and multiple specialized LangGraph agents through an intelligent supervisor routing system.

graph TD
    %% Top Layer - Voice and Routing
    User[👤 User] <-->|Voice| PipeCat[📞 PipeCat Voice Interface]
    PipeCat <-->|Text Query/Response| Supervisor[🧠 Supervisor Agent]
    
    %% Bottom Layer - Multiple LangGraphs
    Supervisor -->|Routes to| MedicalGraph[🏥 Medical LangGraph]
    Supervisor -->|Routes to| LegalGraph[⚖️ Legal LangGraph]
    Supervisor -->|Routes to| MoreGraphs[➕ Add More LangGraphs...]
    
    %% Each Graph can have multiple nodes
    subgraph MedicalGraph[🏥 Medical LangGraph]
        MedNodes[Multiple Nodes: Diagnosis → Treatment → Prescription]
    end
    
    subgraph LegalGraph[⚖️ Legal LangGraph]
        LegalNodes[Multiple Nodes: Research → Analysis → Advice]
    end
    
    %% Styling
    classDef topLayer fill:#e3f2fd,stroke:#1976d2,stroke-width:3px
    classDef graphLayer fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
    classDef nodeLayer fill:#f1f8e9
    
    class User,PipeCAT,Supervisor topLayer
    class MedicalGraph,LegalGraph,MoreGraphs graphLayer
    class MedNodes,LegalNodes nodeLayer

• Voice Interface: Real-time speech-to-text and text-to-speech using PipeCat
• Supervisor Agent: Intelligent routing system that directs queries to appropriate specialized agents
• Specialized Graphs: Domain-specific LangGraph agents (e.g., medical assistant)
• WebRTC Transport: Low-latency voice communication over web protocols

• Real-time voice conversation with AI agents
• Intelligent query routing based on content analysis
• Specialized domain expertise (medical advice, general assistance)
• Interruptible conversations with natural flow
• WebRTC-based low-latency communication
• Extensible architecture for adding new specialized agents

• Python 3.11 or higher
• uv package manager
• OpenAI API key
• Deepgram API key (for speech-to-text)
• Cartesia API key (for text-to-speech)

# On macOS and Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# On Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

git clone <your-repo-url>
cd pipecat-langchain-voice-ai

# Install all dependencies
uv sync

# Or install in development mode
uv sync --dev

Create a .env file in the project root:

OPENAI_API_KEY=your_openai_api_key_here
DEEPGRAM_API_KEY=your_deepgram_api_key_here
CARTESIA_API_KEY=your_cartesia_api_key_here

# Using uv
uv run python 07b-interruptible-langchain.py

# Or activate the virtual environment and run
uv shell
python 07b-interruptible-langchain.py

The application will start a web server (default: http://localhost:7860) where you can interact with the voice AI agent through your browser.

python run.py [bot_file] [--host HOST] [--port PORT] [--verbose]

• bot_file: Path to the bot file (optional, auto-detected)
• --host: Host for HTTP server (default: localhost)
• --port: Port for HTTP server (default: 7860)
• --verbose: Enable verbose logging

1. Create a new LangGraph in graph.py or a separate file
2. Update the supervisor routing logic in supervisor.py
3. Add the new route condition in the supervisor_node function

Edit the system prompt in graph.py:

prompt = ChatPromptTemplate.from_messages([
    ("system", "Your custom medical assistant prompt here"),
    ("human", "{input}")
])

Modify voice settings in 07b-interruptible-langchain.py:

tts = CartesiaTTSService(
    api_key=os.getenv("CARTESIA_API_KEY"),
    voice_id="your_preferred_voice_id",  # Change voice
)

├── supervisor.py              # Main supervisor routing logic
├── graph.py                   # Specialized LangGraph definitions
├── 07b-interruptible-langchain.py  # Main voice AI application
├── run.py                     # Application runner and server setup
├── pyproject.toml            # Project dependencies and metadata
├── .env                      # Environment variables (create this)
└── README.md                 # This file

1. Voice Input: User speaks into the web interface
2. Speech-to-Text: Deepgram converts speech to text
3. Supervisor Routing: The supervisor agent analyzes the query and routes it to the appropriate specialized agent
4. Processing: The selected LangGraph processes the query using OpenAI's GPT model
5. Response Generation: The specialized agent generates a contextual response
6. Text-to-Speech: Cartesia converts the response to natural speech
7. Voice Output: The response is played back to the user

# Run with verbose logging for debugging
uv run python 07b-interruptible-langchain.py --verbose

# Test the web interface
# Navigate to http://localhost:7860 in your browser

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Add tests if applicable
5. Commit your changes (git commit -m 'Add amazing feature')
6. Push to the branch (git push origin feature/amazing-feature)
7. Open a Pull Request

This project is licensed under the BSD 2-Clause License - see the LICENSE file for details.

• PipeCat AI - Real-time voice AI infrastructure
• LangChain - AI application framework
• LangGraph - Multi-agent orchestration

For questions or issues:

1. Check existing Issues
2. Create a new issue with detailed information
3. Include logs and error messages when reporting bugs

Built with ❤️ using PipeCat, LangChain, and LangGraph