WanderWise — Multi-Agent Travel Planner

A multi-agent AI travel assistant built with Google ADK and Gemini. WanderWise orchestrates six specialized sub-agents to deliver a complete travel plan: personalized itinerary, current events, weather forecast, and a packing list — all from a single natural-language query.

Architecture

User Query
    │
    ▼
┌────────────────────────────────────┐
│   wanderwise_coordinator_agent     │  ← routes & orchestrates
└────┬───────┬───────┬───────┬───────┘
     │       │       │       │
     ▼       ▼       ▼       ▼
 itinerary  events  weather  personalizer
  agent     agent   agent    agent
                               │
                               ▼
                         packing_list
                            agent

Agent	Responsibility
itinerary_agent	Creates day-by-day travel itinerary using Google Search
latest_events_agent	Finds current events & festivals via Google Search
weather_agent	Current weather & 5-day forecast via OpenWeatherMap
personalized_itinerary_agent	Merges itinerary with relevant events
packing_list_agent	Generates a packing list from itinerary + weather
wanderwise_coordinator_agent	Top-level orchestrator; routes queries to sub-agents

Prerequisites

• Python 3.10+
• A Google Cloud project with the Vertex AI API enabled, or a Google AI Studio API key
• An OpenWeatherMap API key (free tier is sufficient)

Setup

1. Clone & install dependencies

git clone https://github.com/variang/multi-agent-travel-planner.git
cd multi-agent-travel-planner
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

2. Configure environment variables

Copy the example file and fill in your credentials:

cp .env.example .env

Open .env and fill in the values (see sections below).

---

LLM Connection Options

Option A — Google AI Studio (API Key, simplest)

# .env
GOOGLE_GENAI_USE_VERTEXAI=False
GOOGLE_API_KEY=YOUR_GOOGLE_AI_STUDIO_API_KEY
OPEN_WEATHER_API_KEY=YOUR_OPENWEATHER_KEY

Get a free API key at https://aistudio.google.com/apikey.

---

Option B — Vertex AI via Service Account (recommended for production)

This uses a Google Cloud Service Account JSON key to authenticate with the Vertex AI Gemini API.

Step 1 — Create a Service Account

1. Go to IAM & Admin → Service Accounts in the Cloud Console.
2. Click Create Service Account, give it a name (e.g. wanderwise-sa), and click Create and Continue.
3. Grant the role Vertex AI User (roles/aiplatform.user), then click Done.

Step 2 — Download the JSON key

1. Click your new service account → Keys tab → Add Key → Create new key → JSON.
2. Save the downloaded file, e.g. as service-account.json (keep it outside the repo or add it to .gitignore).

Step 3 — Set environment variables

# .env
GOOGLE_GENAI_USE_VERTEXAI=True
GOOGLE_CLOUD_PROJECT=your-gcp-project-id
GOOGLE_CLOUD_LOCATION=us-central1          # or your preferred region
GOOGLE_APPLICATION_CREDENTIALS=/absolute/path/to/service-account.json
OPEN_WEATHER_API_KEY=YOUR_OPENWEATHER_KEY

Tip: Never commit the JSON key or .env to version control. Both are listed in .gitignore.

Alternative — Application Default Credentials (ADC)

If you're running on GCE / Cloud Run / GKE, or have already run gcloud auth application-default login, you can omit GOOGLE_APPLICATION_CREDENTIALS and ADC is used automatically.

gcloud auth application-default login

---

Observability with Langfuse (optional)

WanderWise integrates with Langfuse for tracing agent runs. Tracing is opt-in — the app works fine without it.

Start a local Langfuse instance

Docker is required. A docker-compose.langfuse.yml is included in the repo.

# 1. Start the Docker daemon (macOS with Colima)
colima start

# 2. Start Langfuse + its Postgres database
docker compose -f docker-compose.langfuse.yml up -d

Langfuse will be available at http://localhost:3000.

Create an account and generate API keys

1. Open http://localhost:3000 and register a new account.
2. Create an organisation and a project.
3. Go to Settings → API Keys and create a new key pair.

Add the keys to your .env

LANGFUSE_SECRET_KEY=sk-lf-...
LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_BASE_URL=http://localhost:3000
LANGFUSE_HOST=http://localhost:3000

Note: Keys generated on Langfuse Cloud will not work against a local instance and vice versa — always use keys from the same instance you are pointing to.

Stopping Langfuse

docker compose -f docker-compose.langfuse.yml down

---

Running the Planner

python main.py

This starts an interactive CLI. Type your travel request and press Enter. Type exit or quit to stop.

To run the canned evaluation test cases:

python eval_test.py

Example queries

"I will be in Milan for 3 days this weekend. I love fashion and food. What should I pack?"
"Plan a 5-day trip to Cairo in September. I love ancient history and local cuisine."
"What events are happening in Munich at the end of July? Any packing tips?"
"What's the current weather in Tokyo?"

Project Structure

multi-agent-travel-planner/
├── agents/
│   ├── __init__.py
│   ├── weather_agent.py         # Current & forecast weather (OpenWeatherMap)
│   ├── itinerary_agent.py       # Day-by-day itinerary (Google Search)
│   ├── events_agent.py          # Events & festivals (Google Search)
│   ├── personalizer_agent.py    # Merges itinerary + events
│   └── packing_list_agent.py    # Packing list generator
├── tools/
│   ├── __init__.py
│   └── weather_tools.py         # Custom OpenWeatherMap tool functions
├── coordinator.py               # Coordinator agent + wrapper tools
├── tracing_utils.py             # Langfuse tracing helpers (optional)
├── main.py                      # Interactive CLI entry point
├── eval_test.py                 # Canned evaluation test cases
├── docker-compose.langfuse.yml  # Local Langfuse + Postgres stack
├── requirements.txt
├── .env.example
└── README.md

Key Dependencies

Package	Purpose
google-adk	Agent Development Kit (agents, runners, sessions)
google-genai	Gemini model access
requests	HTTP calls to OpenWeatherMap
python-dateutil	Flexible date string parsing
python-dotenv	Load .env files
langfuse	Observability & tracing (optional)

License

MIT