Search in plain language
Describe what you're in the mood for — "something cozy to fall asleep to", "80s sci-fi with robots" — and we turn it into real filters. Exact titles stay as fast as ever. 5 searches a day on Free, 100 on PRO, 500 on PRO+.
Until now, finding something without knowing its name meant guessing at keywords and hoping the catalogue used the same words you did. Now you can just say what you want.
Type "something cozy to fall asleep to" or "80s sci-fi with robots" and we read it as filters — genre, decade, rating, quality — and search the catalogue with them.
It never slows down a search that already worked
This was the part we cared most about getting right. Searching for an exact title is the most common thing anyone does here, and it has to stay instant.
So the AI does not run first. Your search runs normally, and only if it comes back empty — or with results that are clearly a poor match — do we spend a moment interpreting the query and try again. If that second attempt finds less than the first, we keep the first. A search can never get worse because the AI got involved.
In practice that means an exact title still answers in a fraction of a second and never touches the model at all. There's also a "Search with AI" button when you want to force it: some phrasings return literal results that are technically matches but not what you meant, and the automatic path won't catch those.
You can see how we read it
Whatever we understood shows up as chips above the results: the genre we picked, the years, the rating floor. If we got something wrong, remove that chip — the rest of your filters stay exactly as they were.
It's not decoration. When an interpretation misses, this is what tells you why the results look the way they do, without making you start over.
How many you get
Every interpreted search costs us a model call, so there's a daily allowance: 5 a day on Free, 100 on PRO and 500 on PRO+. Ordinary searching is unlimited and always has been — this only counts the ones we interpret for you.
Repeating a phrasing someone already searched costs nothing and doesn't touch your allowance. Neither does a search where the model fails to answer: if it times out you get the ordinary results and your count stays where it was.
