Travel Agent

LLM-powered travel planning assistant that combines OpenAI chat models with multiple SerpApi tools (Google Flights, Google Hotels, Google Maps/Local). The agent reasons about user constraints (budget, luxury, flexible dates, etc.), asks clarifying questions, runs searches concurrently, and returns a well-cited answer. It includes an interactive CLI chat for discovery and planning, plus a Python class for programmatic use.

---

Features

• Context-aware reasoning about constraints (budget, luxury, season)
• Multi-tool integration via SerpApi (Flights, Hotels, Maps, Web search)
• Chat-based CLI for discovery & exploration
• Footnote-style citations with source links (tools return structured JSON with URLs to support citations)
• JSON trace output for debugging or auditing

---

Requirements

• Python 3.9+
• Environment variables:
  • OPENAI_API_KEY
  • SERPAPI_API_KEY

---

Install

git clone https://github.com/your-org/travel-planning-agent
cd travel-planning-agent
pip install -r requirements.txt

---

Setup API Keys

Export in your shell (recommended):

export OPENAI_API_KEY="sk-..."
export SERPAPI_API_KEY="..."

Or create a .env file and source it in your shell before running:

source .env

---

Usage

1) Chat mode (interactive)

python travel_planning_agent.py

2) Single-query mode

python travel_planning_agent.py -q "Plan a 5-day luxury trip to Paris in July."

3) With debug logs

python travel_planning_agent.py -q "Find cheap flights to Tokyo in May" --debug

---

CLI Options

• -q, --query : Initial travel query (if omitted → chat mode)
• -m, --model : Model (default: gpt-4o, choices: gpt-4o, gpt-4o-mini)
• -n, --topn : Number of results per search (default: 10)
• -o, --outfile : Write full JSON trace to file
• -d, --debug : Enable debug logs

---

Behavior

• Interprets ambiguous dates as future travel:
  • If the current date is past a mentioned month (e.g., today is August and user says "in May"), assume next year's May.
  • For vague phrasing like "mid-May", probe multiple nearby dates (e.g., 13–17 May) rather than a single exact day.
• Assumes reasonable defaults when details are missing (and surfaces them):
  • Economy cabin, up to 1 stop, 2 adults, flexible +/- 3 days; adjust if tone/budget suggests otherwise.
• Infers defaults based on query tone:
  • "Luxury" → business/first flights, 5-star hotels
  • "$500 budget" → economy flights, value hotels
• Asks clarifying questions if info missing (dates, origin, budget) and provides multiple options when uncertain
• Outputs a concise, actionable plan with links & footnote-style citations

Tooling notes

• Flights tool requires IATA airport codes:
  • The search_flights tool expects 3-letter IATA codes for departure and destination (e.g., WAW, HAM).
  • The agent should first call search_web to find/verify the IATA codes (e.g., search: "Warsaw IATA code", "Hamburg IATA code"), then call search_flights with those codes.
  • If non-IATA inputs are provided, the tool returns a structured error indicating that IATA codes are required.
• Hotels and Places tools accept free-text destination queries (cities, neighborhoods) as before.

---

Project Structure

travel-planning-agent/
├─ travel_planning_agent.py # CLI entrypoint + TravelAgent class
├─ requirements.txt
├─ README.md
└─ LICENSE

---

Example Queries

• Budget trip: I want a $500 budget trip to Mexico City in November
• Discovery: I want a summer beach vacation
• Luxury honeymoon: Plan a luxury honeymoon in Bali in December