OpenAI has open sourced a new multi-agent customer service demo on Github, showing how to build domain-specific AI proxy using its proxy SDK. This project – titled openai-cs-agents-demo– Modeling airline customer service chatbots that can handle a series of travel-related queries by dynamically routing requests to dedicated agents. The system is built by the Python backend and Next.js Frontend, which can provide both a functional dialogue interface and visual traces for proxy handover and guardrail activation.
The architecture is divided into two main components. The Python backend uses the proxy SDK to handle proxy orchestration, while Next.js Frontend provides interactive visualization of the chat interface and proxy transformation. This setting provides transparency for decision-making and authorization processes when a proxy classifies, responds to, or rejects user queries. The demo is with several key agents: Classification Agent, Seat Booking Agent, Flight Status Agent, Cancel Agent and FAQ Agent. These are all configured with special instructions and tools to implement their specific subtasks.
When a user enters a request (such as “Change My Seat” or “Cancel My Flight”), the Triage proxy processes the input to determine the intent and assigns the query to the appropriate downstream proxy. For example, a reservation change request will be routed to a seat reservation agent that can verify the confirmation number, provide seat map selection, and finalize seat change. If the request is cancelled, the system will hand over to the cancel agent, which follows a structured process to confirm and perform the cancellation. The demo also includes a real-time flight inquiry flight status agent and a FAQ agent that answers general questions about baggage policy or aircraft type.
The key advantage of this system is that it integrates security and relevance. The demo has two: the relevant guardrail and a jailbreak guardrail. Related guardrails filter topic queries, such as rejecting tips like “Write me a poem about strawberries”. Jailbreak guardrail blocks attempt to circumvent system boundaries or manipulate proxy behavior, such as requiring the model to reveal its internal instructions. When either type of guardrail is triggered, the system highlights it in the tracking and sends a structured error message to the user.
The agent SDK itself is the backbone of orchestration. Each agent is defined as a composable unit with timely templates, tool access, switching logic and output modes. The SDK handles chain proxy via “handover”, supports real-time tracking, and allows developers to use guardrails to implement input/output constraints. The framework is the same as Openai’s internal experiments and is powered by tools and reasoning agents, but is now exposed in an educational and scalable format.
Developers can run the demo locally by starting a Python backend server with Uvicorn and using a single boot frontend npm run dev Order. The entire system is configurable – developers can insert new agents, define their own task routing policies, and implement custom guardrails. This demonstration features completely transparent tips, decision making and tracking logs, providing a practical basis for real-world conversational AI systems in customer support or other enterprise domains.
By releasing this reference implementation, OpenAI provides a tangible example of how multi-agent coordination, tool usage, and security checking can be combined into a powerful service experience. It is especially valuable for developers looking to understand the anatomy of the agent system and how to build modular, controllable AI workflows that are both transparent and productive.
Check Github page. All credits for this study are to the researchers on the project. Also, please feel free to follow us twitter And don’t forget to join us 100K+ ml reddit And subscribe Our newsletter.
Asif Razzaq is CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, ASIF is committed to harnessing the potential of artificial intelligence to achieve social benefits. His recent effort is to launch Marktechpost, an artificial intelligence media platform that has an in-depth coverage of machine learning and deep learning news that can sound both technically, both through technical voices and be understood by a wide audience. The platform has over 2 million views per month, demonstrating its popularity among its audience.
