Experimental Seat Booking Agent

What is Experimental Seat Booking Agent?

An experimental seat booking agent implementation that combines OpenAI Agents SDK, Autogen extensions, and MCP (Model Context Protocol) tools support. It runs on Linux and provides a Chainlit web chat interface for self-service airline seat reservations.

How to use Experimental Seat Booking Agent?

Set up a Python virtual environment, install dependencies from requirements.txt, and launch the application by running chainlit run app.py from the project root.

Key features of Experimental Seat Booking Agent

Triage agent answers general inquiries using tools
Routes customer questions to specialized agents
Retrieves user preferences on seating, airlines, and food
Uses agents as tools for question routing
Uses MCP server tools for web fetch and file reading
QA agent handles knowledge about planes, flights, and onboarding
Seat booking agent specialized for plane seat reservations
Web chat interface for user self-service

Use cases of Experimental Seat Booking Agent

Self-service airline seat reservation kiosk for comprehensive customer support
Handling reservations, inquiries, and preference retrieval through a call-center-like interface
Routing passenger questions to specialized agents (QA, seat booking) for efficient service
Demonstrating integration of OpenAI Agents SDK, Autogen, and MCP tools in a travel workflow

FAQ from Experimental Seat Booking Agent

What is the purpose of this agent?

It is designed to provide comprehensive customer support at an airline self-service kiosk, handling tasks like reservations, general inquiries, and passenger assistance.

What technologies does it use?

The stack includes Linux, OpenAI Agents SDK, Autogen 0.4 with extensions (MCP Tools, OpenAI, Ollama), Chainlit UI, and Ollama running local Qwen2.5 models.

Does it support MCP servers?

Yes, the triage agent uses MCP servers as tools to perform web fetch content and read file system operations.

How is the project licensed?

The project is licensed under the MIT License.

Are there any known limitations?

The agent is experimental; planned improvements include saving seat history, implementing RAG for the FAQ agent, improving memory storage, and adding MCP tools to invoke an actual booking service API.