Semantic Search

How It Works

Traditional search finds exact keyword matches. Semantic search understands the meaning of your query and finds relevant passages even when the exact words don’t match.

ReadAny uses a hybrid retrieval approach combining:

• Vector similarity — Find passages with similar meaning using embeddings
• BM25 — Traditional keyword matching for precision

Vectorizing a Book

Before using semantic search, you need to vectorize the book:

1. Open a book
2. Click the vectorize button in the sidebar (or it may start automatically)
3. Wait for the process to complete — this runs locally using your CPU

The vectorization creates a local embedding index stored on your device. This is a one-time process per book.

Using Semantic Search

1. Open the Search panel in the sidebar
2. Type a natural language query (e.g., “the protagonist’s childhood memories”)
3. Results are ranked by semantic relevance
4. Click any result to jump to that passage

Embedding Model

ReadAny uses Transformers.js to run embedding models locally in the browser. No data leaves your device.

You can configure the embedding model in Settings → Vector Model.