AI & Agents · Intermediate
LLM Trip Planner With Live Tools
Build an agent that plans a trip by calling real flight, weather, and maps APIs through function calling and streaming the itinerary.
You build a conversational trip-planning agent that orchestrates real external APIs through LLM function calling: Amadeus for flights, Open-Meteo for forecasts, and Mapbox for route geometry. The agent receives a natural-language trip request, decides which tools to invoke and in what order, then streams a structured day-by-day itinerary back to the user. This project teaches you exactly how production agentic systems work, because every piece of data the model uses comes from a live API call rather than its training weights.
What you build
- Natural-language trip requests parsed and fulfilled by an LLM agent without hand-coded routing logic
- Real flight search via Amadeus API returning actual routes, prices, and departure times
- Live 7-day weather forecast per destination from Open-Meteo, injected into itinerary recommendations
- Mapbox route geometry rendered on an interactive map showing the day-by-day travel path
- Streamed itinerary output so users see results as the agent assembles them, not after a long wait
- Multi-turn conversation so users can refine the plan (change dates, add stops, swap flights)
What it teaches
- LLM tool calling and multi-step agentic loops using the Vercel AI SDK `streamText` with `maxSteps`
- Designing Zod-typed tool schemas that give the model precise, safe interfaces to external APIs
- Streaming partial results to the client so UI stays responsive during multi-tool orchestration
- Integrating third-party REST APIs (Amadeus, Open-Meteo, Mapbox) as agent capabilities rather than direct UI calls
- Handling tool execution errors gracefully so the model can retry or fall back without crashing the session
- Rendering geospatial data from an LLM-driven tool call onto an interactive map in a React app
How it works
- 1
User Prompt
- Natural-language trip request
- Budget + date constraints
↓ parsed by
- 2
LLM Agent
- Tool selection
- Parameter extraction
- Zod validation
↓ calls
- 3
External APIs
- Amadeus flights
- Open-Meteo weather
- Mapbox route
↓ returns JSON
- 4
Tool Results
- Flight options
- Forecast data
- Route geometry
↓ fed back to
- 5
Itinerary Generation
- Model synthesizes plan
- Streamed tokens
↓ renders on
- 6
Client UI
- Streamed itinerary text
- Interactive map
Sign in to open the build guide
Free account. Get the step-by-step build and every resource link.
Take it further
- Add a budget-optimization loop where the agent re-calls `searchFlights` with adjusted dates if the first result exceeds the user's stated budget, demonstrating multi-step planning with feedback.
- Persist itineraries to a database and add a share link so users can send a read-only view of their AI-generated plan to travel companions.
- Replace single-city weather with per-day forecasts keyed to each destination stop, then have the model adjust outdoor activity recommendations based on rain probability.
