Three films, explained.
Flick is a decision tool, not a streaming service or a database. Here’s the whole pipeline — and where we deliberately don’t use AI.
We read your request
Your sentence — “something tense, short, and not too gory” — becomes structured constraints: moods, genres, runtime, language, who’s watching, and what to avoid. A rules-based reader handles this by default. When an AI key is configured, a language model does the reading instead, but it can only fill in a fixed vocabulary; anything outside it is discarded.
Rules choose the films
Hard limits first — runtime, language, content limits, who’s watching — remove anything that doesn’t fit. The survivors are scored on genre, mood, keywords, similarity to films you name, and your taste profile. The same request always produces the same ranking, so you can trust it and we can test it.
Exactly three, on purpose
The first pick is the closest match. The next two are chosen to differ — another era, language, length or mood — so you’re comparing real alternatives, not three near-duplicates. Each says how it differs.
Every reason is checkable
Reasons are written from fields that actually matched: “runs 106 minutes — within your 120 limit”, “original language: Korean”. If we can’t back a claim with data, we don’t make it. There are no star ratings and no streaming-availability claims, because we don’t have a verified source for them yet.
It learns, on your device
Love, save, “seen it”, or “not for me” — each nudges future picks toward or away from genres, moods and directors. Seen and rejected films are never suggested again. The catalog holds 1,888 films from 1980 to 2026 — 85 hand-checked with editorial notes, the rest with facts traced to Wikidata. Only the hand-checked ones can satisfy a content limit.