HazelJS powers travel planner with TypeScript

HazelJS released a full-stack TypeScript travel-itinerary planner that demonstrates multi-agent orchestration, retrieval-augmented generation, and production-grade resilience features [Dev.to]. The repository ships a monorepo with a src/agents folder containing five agents: TravelIntakeAgent, DestinationResearchAgent, ItineraryBuilderAgent, LogisticsAgent, and TravelAdvisorAgent. Each agent is declared with the @Agent decorator, and the orchestrator uses @Delegate to route tasks. The DestinationResearchAgent enables RAG with ragTopK: 3 and a MemoryVectorStore backed by a local embedding provider [Dev.to]. Runtime configuration in AppModule sets rateLimitPerMinute: 120, enables retries, circuit breakers, and metrics, and exposes an InspectorModule at /__hazel for live traces [Dev.to]. The project includes a package.json with a build script and curl examples that exercise the intake, destination search, and supervisor endpoints. The codebase also shows how to swap the LLM provider—TravelPlanningLocalLLMProvider can be replaced with OpenAI, Anthropic, or Google Generative AI with a single import change. HazelJS's built-in retry, circuit-breaker, and rate-limiting layers mean engineers can ship a reliable AI service without writing custom middleware [Dev.to]. By abstracting the LLM behind a provider interface, the same agent code can run on OpenAI, Anthropic, or Google models. The HazelJS Inspector streams tool-call logs, step counts, and latency metrics for each agent execution [Dev.to].