A multi-agent trip planning system built with CrewAI for a hackathon. Five specialised AI agents collaborate to research destinations, find flights and accommodations, and build day-by-day itineraries.

• CrewAI Multi-Agent Orchestration - 5 agents that share context and collaborate
• Multi-Provider LLM Support - OpenAI, Google Gemini, and Anthropic Claude
• Real-Time Progress Streaming - SSE endpoint shows each agent's progress live
• Tool-Equipped Agents - web search (Tavily/Serper), flight search, accommodation search, city info
• Multi-City Planning - automatically plans multi-city routes for country-level destinations
• Preference-Aware - respects dietary restrictions, interests, and budget level
• Mock Data APIs - flights and accommodations work without external API keys

Agents pass context through CrewAI's task dependency system - the ItineraryPlanner receives output from all four previous agents to create a cohesive plan.

cd HackEurope26
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Copy the example env file and set your API key:

cp .env.example .env

Then edit .env and set one of these depending on your provider:

Default models per provider:

• OpenAI: gpt-4o-mini
• Gemini: gemini-2.0-flash
• Claude: claude-sonnet-4-20250514

Override with LLM_MODEL=your-model-name if needed.

Option A - One command:

bash start.sh

Option B - Manual (two terminals):

Terminal 1 (backend):

source .venv/bin/activate
python main.py
# API at http://localhost:8000

Terminal 2 (frontend):

source .venv/bin/activate
cd streamlit_app
streamlit run app.py
# UI at http://localhost:8501

1. Open http://localhost:8501
2. Register with any email/password
3. Create a trip (try "Japan" for multi-city or "Tokyo" for single-city)
4. Click "Start Planning" and watch the agents work in real-time

Yes, the system works fully with mock data and requires NO external API keys beyond an LLM key. Here's what happens:

Minimum to run: Just one LLM API key (OPENAI_API_KEY, GEMINI_API_KEY, or ANTHROPIC_API_KEY).

The flight search, accommodation search, and city info tools all use local mock data generators - no Skyscanner/Airbnb/Google Maps keys needed.

For better destination research, set a web search key:

# Tavily (preferred, free tier available)
TAVILY_API_KEY=tvly-...

# OR Serper (alternative)
SERPER_API_KEY=...

Without these, the DestinationResearcher agent still works - it just relies on the LLM's training data instead of live search results.

HackEurope26/
├── agents.py              # CrewAI agents, tools, tasks, and crew orchestration
├── main.py                # FastAPI backend with SSE streaming
├── database.py            # SQLAlchemy models (User, Trip, Flight, etc.)
├── mock_data.py           # Mock flight/accommodation/city data generators
├── requirements.txt       # Python dependencies
├── .env.example           # Environment variable template
├── start.sh               # One-command launcher
├── streamlit_app/
│   └── app.py             # Streamlit frontend
└── docs/                  # Design documentation

• "Japan" triggers multi-city planning (Tokyo, Kyoto, Osaka)
• "Paris" triggers single-city planning
• Watch the SSE stream to see agents hand off context to each other
• The /health endpoint shows which LLM provider is active

• Flight and accommodation data is mock (simulated APIs)
• No real booking integration
• SQLite database (not production-ready)
• Session-based auth (simplified for hackathon)

Users have trip credits — each credit allows planning one AI-powered trip. Credits are displayed in the navbar/banner on every page and can be purchased via Stripe Checkout.

New users start with 3 free credits.

1. Create a Stripe account at dashboard.stripe.com — you'll be in test mode by default.

2. Copy your test secret key from Developers → API keys (starts with sk_test_...) and add to .env:

   STRIPE_SECRET_KEY=sk_test_xxxxxxxxxxxxxx
   FRONTEND_URL=http://localhost:5173

3. Set up local webhook forwarding with the Stripe CLI:

   # Install (Linux)
   curl -s https://packages.stripe.dev/api/security/keypair/stripe-cli-gpg/public | gpg --dearmor | sudo tee /usr/share/keyrings/stripe.gpg
   echo "deb [signed-by=/usr/share/keyrings/stripe.gpg] https://packages.stripe.dev/stripe-cli-debian-local stable main" | sudo tee /etc/apt/sources.list.d/stripe.list
   sudo apt update && sudo apt install stripe
   
   # macOS: brew install stripe/stripe-cli/stripe
   
   # Login & forward
   stripe login
   stripe listen --forward-to localhost:8000/credits/webhook

   Copy the webhook signing secret (whsec_...) and add to .env:

   STRIPE_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxx

4. Test a purchase — use Stripe's test card 4242 4242 4242 4242 with any future expiry and any CVC.

If STRIPE_SECRET_KEY is not set, the checkout endpoint grants credits directly as a hackathon fallback — no real payment is needed to demo the full flow.

On the landing page destination input:

• Type AddCredits and submit → adds 5 credits (must be logged in)
• Type RemoveCredits and submit → removes 5 credits (capped at 0)