How Search Works
Virza’s search is not a keyword box. It’s a research-aware retrieval system that understands what you mean, adapts to how you ask, and learns from your workspace context.
Press ⌘K (Mac) or Ctrl+K (Windows/Linux) to open search from anywhere in the app.
Three search scopes
The search page has three tabs, each designed for a different research workflow:
| Tab | Label in app | What it searches | Best for |
|---|---|---|---|
| All | All | Your library + external research databases, blended | Broad exploration, finding gaps, discovering new papers alongside your existing research |
| My Library | My Library | Only documents, notes, and collections in your workspace | Finding specific papers, cross-referencing, focused analysis |
| Discover | Discover | 600M+ external papers across Semantic Scholar, OpenAlex, PubMed, ArXiv, Crossref, and Exa | Literature discovery, finding papers you don’t have yet, staying current |
My Library
Searches every document, note, collection, and citation in your workspace. Results are ranked by a multi-signal scoring system that considers text relevance, semantic similarity, your workspace’s research context, and recency.
All (smart search)
Combines your library results with a curated selection of external papers. Internal results are prioritized, and external papers appear alongside your library matches, giving you a view of what exists beyond your collection. Papers already in your library are marked as “Saved.”
Discover
Searches across major academic databases without touching your library. Found something interesting? Click Save to Library to import it directly into your workspace.
Discover searches are not metered against your plan’s usage limits. They are rate-limited to 10 requests per minute to ensure fair use.
How queries are understood
Virza automatically detects what kind of question you’re asking and adapts its retrieval strategy:
| Query type | Example | What happens |
|---|---|---|
| Keyword | BERT attention mechanism | Exact lexical matching prioritized. Fast, precise results for known terms. |
| Question | How does attention improve NLP performance? | Semantic understanding activated. Virza generates a hypothetical answer internally to improve retrieval accuracy (a technique called HyDE). |
| Claim | Does caffeine improve cognitive performance? | Evidence-seeking mode. Looks for supporting and contradicting evidence across your library. |
You don’t need to choose a mode. Virza detects the intent automatically based on your query structure. You can also override by selecting a mode from the search options.
How results are ranked
Results are scored using seven signals, weighted dynamically based on query type:
- Cross-encoder reranking (40%): a neural model reads your query and each result together to assess deep relevance
- Semantic similarity (20–40%): how close the meaning of your query is to the document content (boosted for question queries)
- Lexical match (10–60%): traditional keyword overlap (boosted for keyword queries, where it becomes the primary signal)
- Recency: newer papers receive a slight preference, with a 5-year half-life exponential decay
- Research alignment: papers that match your workspace’s research focus are boosted; papers divergent from your research trajectory receive a slight penalty
- Methodology relevance: papers using similar methods to your existing research
- Citation overlap: papers that cite or are cited by documents in your library
Results arrive progressively. You’ll see initial keyword results in under 100 milliseconds, with ranking improving over the next few hundred milliseconds as semantic analysis and neural reranking complete.
What’s searchable
Virza searches across all indexed content in your workspace:
- Document titles, authors, abstracts, and full body text
- Notes content
- Collection names and descriptions
- Citation records
- Extracted table content and figure descriptions
Filters
Combine search with filters to narrow results. Available filters:
| Filter | What it does |
|---|---|
| Collection | Limit search to a specific collection |
| Publication year | Filter by publication date range |
| Methodology | Filter by research methodology tags |
| Geography | Filter by geographic focus |
| Journal | Filter by journal or publication venue |
| Has tables/figures | Only show documents with extracted tables or figures |
See Advanced Filters for details on combining filters.
When search quality degrades
If part of the search infrastructure is temporarily unavailable, Virza degrades gracefully rather than failing:
| Situation | What happens | Impact |
|---|---|---|
| Semantic search unavailable | Falls back to keyword-only ranking | Results based on exact text matches; conceptually related papers may be missed |
| Reranking unavailable | Uses statistical fusion ranking | Slightly less precise result ordering |
| All search infrastructure slow | Returns best results within 4 seconds | May see fewer results than usual |
You’ll never see an error page. Virza always returns the best results it can with the available infrastructure. See Failure Modes for the complete degradation reference.