Planning a trip to Rome in April. 4 people, flying from Poznan or Wroclaw. 8+ route/date combinations to track, prices change daily.

I set up an AI agent on my Raspberry Pi to do it automatically.

Setup

OpenClaw is an open-source framework for running AI agents. It connects to Discord, has web access, file system, and a built-in cron scheduler.

  • Hardware: Raspberry Pi 5, Debian Bookworm, arm64
  • Agent: OpenClaw 2026.2.13
  • Models: Gemini Flash (daily light tasks) + Claude Opus (heavier analysis like flights)
  • Interface: Discord DM

The cron job

OpenClaw’s scheduler runs my flights-poznan-wroclaw-rome job daily at 9:00 CET:

Schedule: 0 9 * * * (Europe/Warsaw)
Model: Claude Opus
Session: isolated

The agent gets a prompt with airports, dates, and formatting rules. It fetches Ryanair’s /api/booking/v4/ endpoint directly - returns JSON with prices and availability, no scraping needed.

Parameters: POZ and WRO to Rome (CIA/FCO), 4 adults, 2 weekends in April, 4 route variants each (same airport return + mixed airport combos).

Output

Every morning on Discord:

LOTY DO RZYMU - APRIL 2026 (4 adults)

TOP 10 - sorted by price:

1. 684 PLN r/t per person - WRO>CIA>WRO | 10-14 Apr
   FR2113 10:45-12:45 | FR2114 15:05-17:10

2. 701 PLN r/t per person - POZ>CIA>WRO | 17-21 Apr
   FR7171 15:50-17:55 | FR2114 15:05-17:10

With booking links, baggage info, price comparisons across return days. The agent caught that Sunday returns are 2-3x more expensive than Tuesday ones, and that there are no Monday return flights to Poznan. Mixed airport routing (fly from POZ, return to WRO) sometimes beats same-airport round trips.

Cost

Student Gemini subscription (free, OAuth) handles daily lightweight tasks like weather checks and heartbeats. Claude comes from my Pro plan which I use daily for coding anyway - the flight job borrows some tokens from that. The Pi draws ~5W.