OpenAI Customer Service Agents Demo

The demo includes several specialized agents that are orchestrated by a Triage Agent:

• Triage Agent: The entry point that routes requests to specialists.
• Flight Information Agent: Provides live status, connection risk, and alternate options.
• Booking & Cancellation Agent: Handles booking, rebooking, or cancelling trips.
• Seat & Special Services Agent: Manages seat assignments and medical/front-row requests.
• FAQ Agent: Answers policy questions (e.g., baggage, compensation, Wi-Fi).
• Refunds and Compensation Agent: Opens cases and issues hotel/meal support after disruptions.

You can run the components either separately or together.

Run the backend independently: Use this if you want to use a separate UI. The backend will be available at http://localhost:8000.

Run the UI & backend simultaneously: Use this for the full demo experience. Running the command from the ui folder will start both the frontend (at http://localhost:3000) and the backend.

# Run backend independently
cd python-backend
python -m uvicorn main:app --reload --port 8000

# Run both UI and backend
cd ui
npm run dev

You can provide your OpenAI API key using one of the following methods:

1. Environment Variable: Run the following command in your terminal:

   export OPENAI_API_KEY=your_api_key

2. .env File: Create an .env file in the python-backend folder. To use this, you must install python-dotenv and add the following to your Python application:

   from dotenv import load_dotenv
   load_dotenv()

The project consists of a Python backend and a Next.js UI. Follow these steps to install dependencies:

Backend Setup (Python): Navigate to the python-backend directory, create a virtual environment, activate it, and install the requirements.

UI Setup (Next.js): Navigate to the ui directory and run npm install.

cd python-backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

cd ../ui
npm install

The snapshot method returns a dictionary representing the current state of a thread, including the thread_id, current_agent, the context, a list of all available agents, and the full history of events and guardrails.

async def snapshot(self, thread_id: Optional[str], context: dict[str, Any]) -> Dict[str, Any]:

The module exports the following core components for interacting with the airline agent system:

Agents:

• triage_agent: Initial entry point for routing queries.
• faq_agent: Handles frequently asked questions.
• flight_information_agent: Provides flight details.
• booking_cancellation_agent: Manages bookings and cancellations.
• seat_special_services_agent: Handles seating and special requests.
• refunds_compensation_agent: Manages refunds and compensation.

Context Types:

• AirlineAgentChatContext: Type for chat-specific context.
• AirlineAgentContext: Base type for agent context.
• create_initial_context: Function to initialize context.
• public_context: Pre-defined public context object.

Use ensure_thread to retrieve an existing thread by its thread_id or create a new one if it does not exist. This method interacts with the underlying store to persist thread metadata.

async def ensure_thread(self, thread_id: Optional[str], context: dict[str, Any]) -> ThreadMetadata:

The respond method is the primary entry point for processing user input. It streams ThreadStreamEvent objects, which include assistant messages and ClientEffectEvent notifications for runner state updates (e.g., runner_bind_thread, runner_state_update, runner_event_delta). It also handles InputGuardrailTripwireTriggered exceptions by returning a refusal message.

async def respond(
        self,
        thread: ThreadMetadata,
        input_user_message: UserMessageItem | None,
        context: dict[str, Any],
    ) -> AsyncIterator[ThreadStreamEvent]: