AI Semantic Search
Search your highlights by meaning. Ask a question in plain language and Screvi finds the passages that answer it, even when they use none of your words.
Updated
On this page
Search by Meaning, Not Keywords
Keyword search only finds what you can already name. Screvi's search understands what you mean. Ask "what did I read about building good habits?" and it returns a highlight about "the power of tiny daily improvements" even though the two share no words.
How to Use It
- Open Search in the web app or the Search tab on mobile.
- Type a question or a topic, at least three characters. Results appear as you type.
- Click a result to open the full highlight, its source, and your notes.
On the web you can switch between one and two columns of results, and the page offers example questions to get you started.
Prompts That Work Well
- "How do I make better decisions under uncertainty?"
- "Quotes about resilience"
- "The compound effect of small habits"
- "Stoic ideas on dealing with criticism"
- "Advice about pricing a product"
- "What have I read about sleep and memory?"
Full sentences and questions tend to beat single words, because they give the search more meaning to match.
Narrowing Down
The search page itself has no filters, by design: it is for open-ended questions. When you want a filtered view, use the Home feed with its type, tag, favorites, and notes filters, or Review > Custom review > Filter Highlights. AI assistants connected through MCP can search with filters for favorites, type, source, and tag.
How It Works
Screvi turns every highlight into an embedding, a numeric fingerprint of its meaning. Your query becomes one too, and Screvi ranks highlights by how close the fingerprints are, blended with a traditional keyword match so exact phrases still win. New highlights are indexed within minutes of import.
Related
- Topic Discovery to see the themes across your library
- Screvi MCP Server to search from Claude and other assistants
Still have questions?