🖇️ Dynamic Travel Planner with AgentStack, CrewAI & Dappier: Real-Time Itinerary Generation Using Multi-Agent AI

This developer-focused cookbook demonstrates how to build and run a local multi-agent itinerary planner using AgentStack, CrewAI, and Dappier, powered by OpenAI and monitored via AgentOps. The setup guides you through generating structured travel itineraries based on real-time weather, live events, and hotel deals for a given city and date.

In this tutorial, you’ll explore:

• AgentStack: A CLI-first framework for rapidly scaffolding AI agent workflows, generating agents, tasks, and tools with seamless CrewAI integration.
• CrewAI: A lightweight multi-agent orchestration engine, perfect for managing sequential or collaborative task execution between agents.
• Dappier: A platform that connects LLMs to real-time, rights-cleared data from trusted sources, specializing in domains like web search, finance, and travel. It delivers enriched, prompt-ready data, empowering AI with verified and up-to-date information for diverse applications.
• OpenAI: A leading provider of advanced AI models capable of natural language understanding, contextual reasoning, and content generation. It enables intelligent, human-like interactions and supports a wide range of applications across various domains.
• AgentOps: Track and analyze the running of CrewAI agents with full visibility into cost, token usage, execution traces, and replays.

This setup not only demonstrates a practical application of AI-driven travel planning but also provides a flexible framework that can be adapted to other real-world scenarios requiring real-time data integration from Dappier, multi-agent collaboration, and contextual reasoning.

📦 Project Initialization and Setup

To get started, we’ll use the agentstack CLI to scaffold a fully functional multi-agent project. This command generates the base project structure, config files, virtual environment, and a ready-to-customize crew.py file with support for tools like Dappier and frameworks like CrewAI.

Step 1: Initialize the Project

Run the following command in your terminal:

agentstack init dynamic_travel_planner

This will generate a folder named dynamic_travel_planner with the complete project structure.

Step 2: Move Into the Project

cd dynamic_travel_planner

Step 3: Activate the Virtual Environment

Activate the virtual environment created by agentstack:

source .venv/bin/activate

If you’re using a Windows terminal, use:

.venv\Scripts\activate

✅ You now have a fully bootstrapped multi-agent AI project using AgentStack and CrewAI.

🔑 Setting Up API Keys

To enable real-time data access and AI reasoning, you’ll need API keys for the following services:

• OpenAI – for LLM-powered reasoning and summarization
• Dappier – for real-time stock market data and web search
• AgentOps – for run monitoring and debugging

These keys are stored in the .env file created during project initialization.

Step 1: Open the .env file

Inside your project root, open .env and update it with your API keys:

# OpenAI
OPENAI_API_KEY=your_openai_key_here

# Dappier
DAPPIER_API_KEY=your_dappier_key_here

# AgentOps (Optional but recommended)
AGENTOPS_API_KEY=your_agentops_key_here

You can get your keys here:

• 🔑 Get your OpenAI API Key here
• 🔑 Get your Dappier API Key here — free credits available
• 🔑 Get your AgentOps API Key here - enables run tracking and replay

🧪 Make sure to keep these keys private. Do not commit .env to version control.

⚙️ Installing Dependencies

After initializing the project and configuring your API keys, install the required dependencies to ensure everything runs smoothly.

The key packages used in this project are:

• crewai – Multi-agent execution and orchestration
• agentstack – CLI toolchain for generating agents, tasks, and crews
• dappier – Real-time data access layer for tools like stock search and news
• openai – LLM access to GPT-4o and other OpenAI models (automatically included)

Step 1: Sync All Dependencies from pyproject.toml

If you’re using the pre-generated project setup from agentstack init, run:

uv lock
uv sync

• uv lock will generate the uv.lock file based on your pyproject.toml
• uv sync will install all dependencies into the virtual environment

Step 2: Add or Upgrade Individual Packages

You need to upgrade packages manually, use:

uv remove "agentstack[crewai]"
uv add crewai
uv add agentstack

✅ You now have all the required dependencies installed locally and locked for reproducible agent execution.

👤 Creating Agents

Now that your environment is ready, let’s generate the agents that will power the travel planning workflow. These agents are created using the AgentStack CLI and are defined declaratively in agents.yaml.

Step 1: Generate the Agents

You’ll use the following command to scaffold each agent:

agentstack generate agent

You’ll be prompted to enter the agent’s name, role, goal, backstory, and model. Repeat this step for each agent in your system.

Here are the agents created for this project:

🗺️ travel_planner

agentstack g a travel_planner \
  --role="A real-time travel assistant that collects weather data, events, and hotel information using the Dappier real-time web search tool." \
  --goal="To generate personalized multi-day itineraries based on current weather, events, and travel deals." \
  --backstory="Trained to support personalized trip planning, this agent uses Dappier's real-time search to retrieve local weather, events, and hotel offers for a target city and date. It creates structured itineraries tailored to weather conditions and event availability." \
  --llm=openai/gpt-4o

📋 itinerary_reporter

agentstack g a itinerary_reporter \
  --role="A travel itinerary formatter that compiles data into a complete multi-day markdown itinerary." \
  --goal="To generate a markdown-formatted travel itinerary based on outputs from the travel planner." \
  --backstory="This agent specializes in transforming weather and event-based trip data into user-friendly markdown plans. It includes timing, local tips, daily highlights, and weather context for each activity." \
  --llm=openai/gpt-4o

✅ Creating Tasks

Each task defines a specific responsibility to be performed by one of the agents. In this project, tasks are tightly scoped and executed sequentially by CrewAI, allowing agents to collaborate and generate a complete travel itinerary using real-time data.

Tasks are defined declaratively in tasks.yaml and are created using the AgentStack CLI.

Step 1: Generate Tasks Using the CLI

Run the following command to create a new task:

agentstack generate task

You’ll be prompted to provide:

• task_name (required)
• --description – a detailed instruction including {city}, {travel_date}, {num_days}, and {timestamp}
• --expected_output – structured data, tables, or markdown expected from the task
• --agent – the agent responsible for this task

Here are the tasks used in this project:

📅 determine_travel_period

agentstack g t determine_travel_period \
  --description="As of {timestamp}, use real-time web search to determine the current date and calculate the travel period based on the user's selected travel date: {travel_date} and trip length: {num_days} days in {city}." \
  --expected_output="A structured summary showing current date, travel start and end dates for the selected trip." \
  --agent=travel_planner

🌤 fetch_weather_forecast

agentstack g t fetch_weather_forecast \
  --description="As of {timestamp}, use real-time web search to retrieve the multi-day weather forecast for {city} during the selected travel period starting on {travel_date} for {num_days} days." \
  --expected_output="Daily weather forecast for {city} including temperature, conditions, and any alerts during the trip." \
  --agent=travel_planner

🎉 fetch_live_events

agentstack g t fetch_live_events \
  --description="As of {timestamp}, use real-time web search to find live events happening in {city} during the travel period starting on {travel_date} for {num_days} days. Include concerts, festivals, sports, and cultural events." \
  --expected_output="A list of events with names, dates, descriptions, and links where applicable for each day of the trip." \
  --agent=travel_planner

🏨 fetch_hotel_deals

agentstack g t fetch_hotel_deals \
  --description="As of {timestamp}, use real-time web search to find hotel deals in {city} during the travel period starting on {travel_date} for {num_days} days. Include price estimates and booking links." \
  --expected_output="Top hotel deals with prices, availability, and booking URLs for the selected city and dates." \
  --agent=travel_planner

📆 generate_travel_itinerary

agentstack g t generate_travel_itinerary \
  --description="As of {timestamp}, compile a detailed {num_days}-day travel itinerary for {city} by combining the outputs of all previous tasks: travel period, weather forecast, live events, and hotel deals. Include day-wise plans with weather-based suggestions, timing, activities, event highlights, hotel info, and booking links." \
  --expected_output="A markdown-formatted itinerary for {city} covering {num_days} days, including: daily schedule, activity suggestions, weather, events, hotel info, and travel tips." \
  --agent=itinerary_reporter

⚠️ The following two fields must be added manually to the generate_travel_itinerary task inside tasks.yaml, as they are not currently supported via the CLI:

output_file: reports/{city}_travel_itinerary.md
create_directory: true

🪄 All of the above tasks will be executed in sequence using CrewAI when you run the crew.

🛠️ Adding Tools to Agents

Tools enable agents to interact with external services like Dappier. In this project, Dappier provides real-time access to weather data, live events, and hotel deals, powering all the data-gathering tasks.

Add Dappier Tools to the Project

Instead of manually assigning tools one-by-one, you can add the full Dappier toolkit using:

agentstack tools add dappier

This:

• Adds all Dappier tools to the AgentStack project
• Automatically installs the dappier package by updating your pyproject.toml file
• Registers the tools under agentstack.tools["dappier"] for in-code access

📝 Providing Inputs & Setting Timestamps

Before running the crew, you need to define the runtime inputs that agents and tasks will use. The AgentStack project already includes an inputs.yaml file, which is used to inject these inputs when the crew is executed.

In this project, we use three dynamic inputs:

• city: The destination city (e.g., “paris”)
• travel_date: The planned start date of the trip (e.g., “tomorrow”)
• num_days: The number of days for the trip (e.g., 5)
• timestamp: The current UTC time, injected automatically via code

Step 1: Update inputs.yaml

Open the pre-generated inputs.yaml file and set the travel details:

city: paris
travel_date: tomorrow
num_days: 5

You can modify the values to plan your own custom itinerary.

Step 2: Inject a Real-Time Timestamp in main.py

To provide the current timestamp at execution time, update the run() function in main.py:

from datetime import datetime, timezone

def run():
    """
    Run the agent.
    """
    inputs = agentstack.get_inputs()
    inputs["timestamp"] = datetime.now(timezone.utc).isoformat()

    instance.kickoff(inputs=inputs)

This will:

• Dynamically inject the current UTC timestamp into the input dictionary
• Allow all tasks referencing {timestamp} in tasks.yaml to use consistent timing context

⏱️ Timestamped input ensures your travel itinerary is grounded in real-time context.

🚀 Running the Crew

Once your agents, tasks, tools, and inputs are all set up, you’re ready to run the multi-agent crew. The crew will execute each task in sequence, collaborating to generate a fully structured itinerary using real-time data.

Step 1: Run the AgentStack Project

To start the crew execution, run the following command from the project root:

agentstack run

This command:

• Loads your agents from agents.yaml
• Loads your tasks from tasks.yaml
• Injects inputs from inputs.yaml (including the runtime timestamp)
• Executes all tasks sequentially via CrewAI
• Stores the final output (e.g., markdown itinerary) in the path defined in tasks.yaml

✅ You should see terminal output as each agent completes its assigned task.

Step 2: Debug with --debug Mode (Optional)

For detailed execution traces, run with the debug flag:

agentstack run --debug

This enables verbose logging, including:

• Which agent is running
• Which tool is being used
• Real-time function call results
• Intermediate outputs for each task

🧪 Use debug mode to troubleshoot tool usage or model behavior during each step.

Step 3: View the Final Output

After the crew finishes execution, you’ll find the generated itinerary at:

reports/paris_travel_itinerary.md

# Paris 5-Day Travel Itinerary  

**Travel Dates:** May 2, 2025 - May 6, 2025  
**Hotel:** Choose from options like Hôtel Malte - Astotel, Hôtel Maison Mère, or others available with deals. Ensure to book through platforms like Trivago or Expedia for competitive prices and free cancellation offers.  

---

## Day 1: May 2, 2025 - Arrival and Exploration  

- **Weather:** Sunny, high around 83°F (28°C)  
- **Morning:**  
  - **Check-in to Hotel** - Settle into your chosen accommodation from the recommended list. Freshen up and prepare for a day of exploration in beautiful weather.

- **Afternoon:**  
  - **St-Germain Jazz Festival** - Dive into the Parisian music scene with jazz performances across the city. Make sure to check [Event Link] for schedules and venues.
  - **Lunch at a Café** - Enjoy dining al fresco with delicious French cuisine nearby.

- **Evening:**  
  - **Taste of Paris Festival** - Immerse yourself in a gastronomic paradise at the Grand Palais. [Event Link] offers details on participating chefs and menu highlights.

- **Local Tip:** Use public transport for convenient travel across the city; a Navigo card can be beneficial for your stay.

---

## Day 2: May 3, 2025 - Historical Charms and Entertainment  

- **Weather:** Partly cloudy, high around 80°F (27°C)  
- **Morning:**  
  - **Visit the Louvre Museum** - Beat the crowds with an early visit. Don’t miss the Mona Lisa and other world-renowned artworks.

- **Afternoon:**  
  - **Roland Garros (French Open)** - Head to the courts for some thrilling tennis matches. Pre-book your tickets via the [Event Link] to secure a spot.

- **Evening:**  
  - **Candlelight Concerts at Sainte-Chapelle** - Revel in an intimate classical music performance in a mesmerizing Gothic setting. [Event Link] for details.

- **Highlight:** Capture sunset views from the Pont Alexandre III.

---

## Day 3: May 4, 2025 - Art and Culture Extravaganza  

- **Weather:** Mostly sunny, high around 81°F (27°C)  
- **Morning:**  
  - **Street Art Fair in Montmartre** - Discover vibrant murals and meet local artists. Be sure to visit iconic spots like Sacré-Cœur.

- **Afternoon:**  
  - **European Museums Night** - Take advantage of free and extended openings at major museums, including special events at the Louvre. Check [Event Link] for participating venues.

- **Evening:**  
  - **Dinner at Le Consulat in Montmartre** - Experience traditional French bistro vibes with authentic dishes.

- **Travel Tip:** Wear comfortable shoes for walking to fully absorb Montmartre’s artistic ambiance.

---

## Day 4: May 5, 2025 - Adventure and Elegance  

- **Weather:** Chance of rain, high around 78°F (26°C)  
- **Morning:**  
  - **Great Paris Steeplechase** - Enjoy the thrill of horse racing; rain gear might be handy. [Event Link] for detailed info and tickets.

- **Afternoon:**  
  - **Cosplay Festival** - Celebrate pop culture at Paris Expo Porte de Versailles, a must for anime lovers. Check [Event Link] ahead for workshops and entry details.

- **Evening:**  
  - **Dinner in Saint-Germain-des-Prés** - Dine at a cozy eatery exploring culinary delights from modest brasseries to refined Parisian cuisines.

- **Pro Tip:** Be prepared for intermittent showers with a compact umbrella.

---

## Day 5: May 6, 2025 - Farewell Paris  

- **Weather:** Cloudy, high around 76°F (24°C)  
- **Morning:**  
  - **Leisurely Seine River Cruise** - Relax on a boat tour taking in iconic sites like the Eiffel Tower and Notre-Dame.

- **Afternoon:**  
  - **Final Stroll and Shopping** - Explore Le Marais district for unique boutiques and last-minute souvenirs.

- **Evening:**  
  - **Candlelight Concerts** - Cap off your trip with a final magical evening of classical music. Tickets and venue details on [Event Link].

- **Farewell Tip:** Dedicate time for a memorable café stop to enjoy a final indulgent Parisian pastry.

---

**Enjoy your unforgettable journey in Paris!** Make sure to review all event schedules and book your hotel early to embrace Paris at its finest. Bon voyage! 🌟

You can open this file in any markdown viewer or commit it to your workspace.

📄 The final itinerary includes weather, events, hotel deals, and daily plans — all compiled in real time.

📊 Viewing Agent Run Summary in AgentOps

This project integrates with AgentOps to provide full visibility into agent execution, tool calls, and token usage. By setting the AGENTOPS_API_KEY in your .env file, all runs are automatically tracked.

Below is a sample AgentOps run for this project:

• Duration: 04m 12s
• Cost: $0.0718500
• LLM Calls: 29
• Tool Calls: 4
• Tokens: 60,006

You can view the complete execution trace, including tool calls, function arguments, and model responses in AgentOps.

🖇 AgentOps: Replay: https://app.agentops.ai/sessions?trace_id=234cfee99d72a45edf8d4fbd3d825461

Note: The AgentOps link shown above is tied to the AgentOps account. To access your own replay, you must run the crew using your personal AGENTOPS_API_KEY.

🌟 Highlights

This cookbook has guided you through setting up and running a local travel planning workflow using AgentStack, CrewAI, and Dappier. You created a structured multi-agent execution that generates real-time itineraries based on weather, events, and hotel data.

Key tools utilized in this cookbook include:

• AgentStack: A CLI framework to scaffold agents, tasks, and tools with support for local execution and configuration.
• CrewAI: A lightweight multi-agent framework that executes agents in sequential or collaborative workflows.
• OpenAI: A leading provider of AI models capable of language understanding, summarization, and reasoning, used to power each agent’s intelligence.
• Dappier: A platform that connects agents to real-time, rights-cleared data from trusted sources, specializing in domains like web search, finance, and travel.
• AgentOps: A tool to track and analyze agent runs, including replay, cost breakdowns, and prompt histories.

This comprehensive setup allows you to adapt and expand the example for various scenarios requiring advanced data retrieval, multi-agent collaboration, and real-time context integration.