Touri — AI-Powered Travel Planner

Touri is a multi-agent AI travel planning system built with CrewAI and Groq. It orchestrates specialized AI agents to search for real flights, hotels, and build day-by-day itineraries — all through a Streamlit web interface.

---

Features

• Multi-agent architecture — dedicated agents for flights, accommodation, and itinerary planning
• Real data — live flight search via FlightAPI.io, hotel search via Booking.com (RapidAPI), weather via OpenWeatherMap
• Currency conversion — automatic price conversion to 150+ world currencies via ExchangeRate API
• PDF export — download your complete travel plan as a formatted PDF
• Budget breakdown — visual pie chart of cost distribution across flights, accommodation, and activities
• Weather forecast — 5-day forecast for your destination integrated into the overview

---

Project Structure

Travel Planner/
├── main.py                        # Entry point — validates env and launches Streamlit
├── pyproject.toml                 # Project metadata and dependencies (uv/pip)
├── .env                           # Your API keys (not committed)
├── .env.example                   # Template for required environment variables
│
├── app/
│   └── streamlit_app.py           # Full Streamlit UI — sidebar inputs, tabs, PDF download
│
└── src/
    ├── agents/
    │   ├── base_agent.py          # Base class — Groq LLM setup, retry logic, agent factory
    │   ├── flight_agent.py        # Searches for flights using FlightAPITool
    │   ├── accommodation_agent.py # Searches for hotels using BookingComTool
    │   └── itinerary_agent.py     # Builds day-by-day plans using WeatherTool
    │
    ├── crew/
    │   └── travel_crew.py         # Orchestrates agents, runs tasks sequentially,
    │                              # handles rate limits, currency conversion, plan assembly
    │
    ├── models/
    │   └── travel_model.py        # Pydantic models — TravelPlan, FlightOption,
    │                              # AccommodationOption, DailyItinerary, Activity, etc.
    │
    ├── tools/
    │   ├── flight_tool.py         # FlightAPI.io integration — one-way and round-trip search
    │   ├── booking_tool.py        # Booking.com RapidAPI — two-step city→hotel search
    │   ├── weather_tool.py        # OpenWeatherMap — 5-day forecast by city and date range
    │   └── currency_tool.py       # ExchangeRate API — USD to any currency conversion
    │
    └── utils/
        └── pdf_generator.py       # ReportLab-based PDF generation for travel plans

---

Prerequisites

• Python 3.10–3.13
• uv (recommended) or pip

---

Installation

Using uv (recommended):

uv sync

Using pip:

pip install -e .

---

Configuration

Copy .env.example to .env and fill in your API keys:

cp .env.example .env

# Required
GROQ_API_KEY=your_groq_api_key          # https://console.groq.com/keys

## Tools 
FLIGHT_API_KEY=your_flightapi_key       # https://flightapi.io
RAPIDAPI_KEY=your_rapidapi_key          # https://rapidapi.com (Booking.com API)
OPENWEATHER_API_KEY=your_openweather_key # https://openweathermap.org/api
EXCHANGE_RATE_API_KEY=your_exchange_key  # https://exchangerate-api.com

---

Running

uv run main.py

Then open http://localhost:8501 in your browser.

---

How It Works

1. User fills in departure city, destination, dates, budget, currency, and preferences in the sidebar
2. Three agents run sequentially, each in its own CrewAI crew with a 12-second pause between them to respect Groq's rate limits:
   • FlightAgent → calls FlightAPI.io, returns best flight option
   • AccommodationAgent → calls Booking.com, returns best hotel option
   • ItineraryAgent → calls OpenWeatherMap, builds a day-by-day activity plan
3. Post-processing in TravelCrew._create_travel_plan:
   • Dates are corrected to match the actual trip start date
   • Prices are converted from USD to the user's selected currency via ExchangeRate API
   • Weather forecast is fetched directly and attached to the plan
4. Results are displayed across four tabs: Overview, Flights, Accommodation, Itinerary
5. PDF export is available from the Itinerary tab

---

Models Available

Configured in src/agents/base_agent.py. Switch via the model parameter in TravelCrew:

Key	Model	Notes
llama-70b	llama-3.3-70b-versatile	Default — best quality, 12k TPM free tier
llama-8b	llama-3.1-8b-instant	Faster, 6k TPM free tier
gpt-oss-20b	openai/gpt-oss-20b	Good balance
gpt-oss-120b	openai/gpt-oss-120b	Highest quality
llama-4-scout	meta-llama/llama-4-scout-17b-16e-instruct	Preview
qwen3	qwen/qwen3-32b	Preview

---

Rate Limits

Groq's free tier has tight TPM (tokens per minute) limits. Touri handles this with:

• 12-second sleep between agent tasks
• Exponential backoff retry (up to 5 attempts) on rate limit errors
• LiteLLM configured with num_retries=6

Upgrading to Groq's Dev tier removes these constraints.

---

Tech Stack

Layer	Technology
AI Agents	CrewAI
LLM Provider	Groq (LiteLLM)
UI	Streamlit
Data Models	Pydantic v2
PDF Generation	ReportLab
Charts	Plotly
Package Manager	uv