Why this matters
Complex customer-service scenarios are rarely handled well by a single prompt or model call — they require routing, stateful handoffs, tool use, and safety guardrails. This demo makes those orchestration patterns explicit by showing a small, end-to-end stack that runs multiple specialist agents (triage, flight info, booking, seat services, FAQ, refunds), visualizes routing decisions, and enforces relevance/jailbreak guardrails so you can evaluate both behavior and safety trade-offs quickly.
What Sets It Apart
- Agent orchestration implemented with the OpenAI Agents SDK (Python backend): the repo contains the orchestration logic used to route intents, call tools (mock booking/flight lookups), and hand off between specialists — so you can inspect real agent-to-agent flows instead of just a single-chat example.
- Visual, interactive UI built in Next.js with ChatKit: the frontend lets you watch routing decisions, guardrail triggers, and mock tool calls in a conversation-centric view, which helps surface design and UX issues early.
- Prebuilt specialist agents and demo flows: includes concrete airline-focused flows (seat changes, cancellations, irregular ops with rebooking) and mock itineraries so you can exercise handoffs, compensation logic, and guardrail behavior without external integrations.
- Demo-ready but explicit limitations: the project is designed for prototyping and education (MIT license), not production—mock data and example tools show patterns but aren’t integrated with real airline systems or enterprise authentication.
Who It's For and Tradeoffs
Great fit if you are an engineer or product designer prototyping multi-agent conversational workflows, an educator teaching agent orchestration patterns, or a researcher studying guardrails and tool handoffs. You get concrete examples of routing logic, guardrail behavior, and UI-level observability.
Look elsewhere if you need production-ready connectors, certified data handling/compliance, high-availability deployment patterns, or native integrations with real airline APIs. The demo requires an OpenAI API key and uses mock flight/booking data, so expect to extend tools, storage, and security before production use.
Where it fits
Practically, this repo sits between single-chat demos (prompt examples) and full production platforms: it’s a hands-on reference for implementing agent orchestration patterns and building a conversation UI that surfaces internal agent decisions and safety checks.
