OpenAI Customer Service Agents Demo
https://github.com/openai/openai-cs-agents-demo- Category
- AI Agents
- Rank
- No. 972Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- openai
- GitHub
- 6.6k stars
- Date
About
Reference demo of a customer-service workflow built with the OpenAI Agents SDK — multi-agent handoffs, tools, and a sample frontend.
What it does
An airline support sandbox where conversations move among specialists for flights, bookings, seats, policies, and compensation. The interface shows routing, tool activity, guardrail results, conversation context, and an interactive seat map while users chat.
Why it's ranked here
Worth ten minutes because it makes hidden orchestration behavior visible. Three documented scenarios cover routine requests, policy enforcement, and a disrupted connection that triggers rebooking, seat reassignment, and compensation. Its scope remains clearly demonstrative rather than production-ready.
What's good
The specialist boundaries are concrete, not cosmetic. Handoffs connect distinct booking, flight, seating, policy, and compensation responsibilities. The UI exposes messages, tool calls, outputs, context changes, progress, and guardrail checks, helping engineers understand why a conversation changed course.
Tradeoffs
Conversation threads and attachments live only in process memory, so restarts lose them. Airline records, seat availability, itineraries, and compensation behavior are mocked. Local cross-origin settings target one frontend address, and deployment hardening is left to the adopter.
How to use it well
Use it to learn agent handoffs, prototype airline support flows, or demonstrate orchestration behavior to product and engineering teams. Modify prompts, guardrails, tools, and mock scenarios before adapting the pattern. It does not supply durable storage or production airline integrations.
Technical notes+
python-backend/main.py defines a FastAPI service with /chatkit, /chatkit/state, /chatkit/bootstrap, /chatkit/state/stream, and /health endpoints; streaming uses server-sent events. python-backend/server.py subclasses ChatKitServer, tracks per-thread ConversationState, records handoffs, tool activity, messages, guardrails, and broadcasts deltas. python-backend/memory_store.py implements thread, item, and attachment storage in dictionaries. ui/next.config.mjs proxies ChatKit traffic to port 8000, while ui/app/page.tsx hydrates orchestration state and combines AgentPanel with ChatKitPanel. ui/components/seat-map.tsx renders fixed cabin sections, occupied seats, and exit rows.
Observed
- License
- MIT License
- Languages
- Python backend and TypeScript Next.js UI
- Install surface
- Backend dependencies install from requirements.txt with pip; UI dependencies install with npm
- Interfaces
- FastAPI HTTP API, server-sent event state stream, and browser chat interface
- Runtime services
- Local backend runs on port 8000 and frontend on port 3000
- Storage
- Threads, thread items, and attachments use an in-memory store
- Structure
- Separate backend orchestration service and Next.js visualization interface
Read from README.md, ui/next-env.d.ts, ui/next.config.mjs, ui/postcss.config.mjs, ui/tailwind.config.ts, python-backend/main.py, python-backend/server.py, python-backend/__init__.py, python-backend/memory_store.py, ui/lib/api.ts, ui/app/page.tsx, ui/lib/types.ts, ui/lib/utils.ts, ui/app/layout.tsx, ui/components/seat-map.tsx.
What it can do
Route customer inquiries to appropriate agents
Customer service request or query → Assignment to specialized agent
Transfer conversations between multiple agents
Active conversation and handoff criteria → Seamless agent transition
Execute customer service tools and functions
Customer request requiring tool usage → Tool execution results and responses
Process multi-step customer service workflows
Complex customer service scenario → Structured workflow completion
Demonstrate agent coordination patterns
Customer service use case → Working multi-agent interaction example
Provide sample frontend interface for agent interactions
User interaction through web interface → Visual demonstration of agent capabilities
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.