the manual

01 // how it works

You give us a slack window. We translate that into search URLs across multiple flight aggregators, scrape them in parallel, normalize the results into a single schema, and rank by deal score.

the pipeline

You submit a search — origin (IATA), optional destination, date window (this week / 60d / etc).
We check the cache. Identical searches within 6h return instantly from our database.
We scrape Google Flights first using Firecrawl with a structured JSON extraction schema.
If Google blocks or returns empty, we fall back to Skyscanner — same schema, different URL.
If both fail and you set a destination, we hit Duffel's test-mode API for real airline inventory at sandbox prices (clearly tagged so you don't book by mistake).
Results are normalized into a single shape (origin → dest, date, price in cents, airline, stops, source) and a relative deal score is computed against the result set.
We cache the search for 6h and render the deal grid.

deal score

Within any result set we normalize price linearly: the cheapest deal scores 1.0, the most expensive scores 0.0. This is intentionally simple — we want relative ranking inside a search, not a global "good deal" oracle.

what we don't do (yet)

Book the flight for you — we always link out to the airline / OTA.
Track your card. We never see payment data.
Multi-stop loop optimization (in v1.1).
AI dossiers on every deal (Phase 3 — coming).