A movie buff that only answers from its own shelf of recent films. It finds the most relevant movies with
embeddings computed in your browser, then a free LLM writes the answer from them.
How it works
Embeddings in the browser. Each movie is turned into a vector by all-MiniLM-L6-v2 running on Transformers.js (downloaded once, about 23 MB). No vector database needed.
Retrieval. Your question is embedded the same way, and the closest movies by cosine similarity become the context.
Conversation memory. Follow-ups like "is it scary?" are first rewritten into a standalone question, so retrieval still finds the right film.
Generation. A free OpenRouter model answers using only that context, through AI Valley's serverless proxy.