Hayhooks + Itinerary Agent + Open WebUI

An AI-powered travel itinerary planning system that combines Hayhooks (Haystack AI pipeline serving), a multi-agent itinerary planning system, and Open WebUI for a chat interface.

Features

• Multi-agent architecture: specialized agents for different aspects of travel planning
• Real-time data: live data from Google Maps, route optimization, and Perplexity
• Hierarchical planning: from macro route planning to detailed daily itineraries
• Accommodation optimization: strategic lodging placement for multi-day trips
• Chat interface: Open WebUI for natural-language interaction
• Containerized: one docker compose up to run the whole stack locally

Architecture

Core agents

1. Macro Itinerary Agent (self.agent)
   • Main orchestrator for multi-day travel planning
   • Establishes macro routes and coordinates the other agents
2. Day Itinerary Agent (day_itinerary_agent)
   • Creates detailed single-day itineraries
   • Tool name: daily_itinerary_planning_agent
3. Lodging Agent (lodging_itinerary_agent)
   • Determines optimal accommodation placement across the route
   • Tool name: accommodation_strategy_optimizer

External integrations

• Google Maps API — place search, details, location data
• Optimal Route service — multi-destination route optimization
• Perplexity API — real-time research and local insights

Quick start

Prerequisites

• Docker and Docker Compose
• API keys for Google Maps, Perplexity, and OpenAI

1. Configure environment variables

Copy the example and fill in your keys:

cp .env.example .env

Required:

HAYHOOKS_PIPELINES_DIR=/pipelines
HAYHOOKS_SHOW_TRACEBACKS=true
GOOGLE_MAPS_API_KEY=your_google_maps_api_key   # used by google-maps and optimal-route
PERPLEXITY_API_KEY=your_perplexity_api_key
OPENAI_API_KEY=your_openai_api_key

Optional — enable Langfuse tracing of the Haystack pipeline:

HAYSTACK_CONTENT_TRACING_ENABLED=true
LANGFUSE_SECRET_KEY=your_langfuse_secret_key
LANGFUSE_PUBLIC_KEY=your_langfuse_public_key
LANGFUSE_HOST=https://cloud.langfuse.com

2. Start the stack

docker compose up -d --build

3. Open the UI

Navigate to http://localhost:3000.

Auth is disabled by default for local use — you'll go straight to the chat.

4. Logs / status

docker compose logs -f
docker compose ps

5. Stop

docker compose down

Usage

Start a new chat and describe your trip, for example:

Plan a 5-day trip to Italy focusing on art and cuisine.
I prefer boutique hotels, have a moderate budget, and will be traveling by train.
I want to visit Rome, Florence, and Venice.

Project structure

.
├── pipelines/
│   └── itinerary_agent/
│       ├── pipeline_wrapper.py                 # Multi-agent orchestration
│       ├── macro_itinerary_system_prompt.txt
│       ├── day_itinerary_system_prompt.txt
│       └── lodging_itinerary_system_prompt.txt
├── docker-compose.yml
├── Dockerfile
├── .env.example
└── README.md

Services

Service	Purpose	Network port
open-webui	Chat interface	3000 (host)
hayhooks	Haystack pipeline server	1416 (internal)
google-maps	Google Maps MCP server	8100 (internal)
optimal-route	Route optimization MCP server	8080 (internal)
perplexity	Perplexity MCP server	8080 (internal)

Only open-webui is exposed on the host. The other services communicate over the Docker network.

Configuration

Model

The system uses OpenAI's gpt-4.1 by default. To change it, edit the llm in pipelines/itinerary_agent/pipeline_wrapper.py:

llm = OpenAIChatGenerator(model="o4-mini")  # or your preferred model

Adding new tools

1. Define the MCP toolset in pipeline_wrapper.py
2. Add the tools to the appropriate agent(s)
3. Update the relevant system prompt
4. Rebuild: docker compose up -d --build

Input validation

pipeline_wrapper.py includes a lightweight policy-compliance agent that rejects requests over 10 days / 7 locations or that look like prompt-injection attempts. Remove or relax it in load_validation_system_message() if you don't need it.