Searching Your Library
Beaver uses agentic search. Rather than running one query against one index, it actively explores your library: choosing among several tools, filtering by metadata, and iterating until it has enough to answer you. Knowing how those tools work helps you phrase a question well and understand what Beaver can and can't find.
Finding papers
Topic search
Finds papers that are conceptually related to your query, based on the titles and abstracts in your library. It matches meaning rather than words, so a search for "crime" can surface papers on incarceration, policing, or violence even when they never use the word. This is what Beaver reaches for when you want to explore a topic.
Metadata search
Searches the bibliographic information Zotero stores: authors, years, titles, journals. Use it when you know something concrete about the reference you want. It's also how Beaver chases down a paper it saw cited somewhere else.
Beaver can also simply browse — listing what's in a library, a collection, or under a tag, with no query at all.
Getting answers out of papers
Asking one question of many papers at once
Beaver can send the same question to a whole set of papers — up to 30 at a time — and come back with a short, cited answer for each one. This is the efficient way to compare methods across a literature, screen a reading list for relevance, or pull one specific number out of every paper in a collection.
Answering across many papers at once is part of Plus Tools, included with Plus. See Credits for what it costs.
Searching inside documents
Once Beaver knows which papers matter, it can keyword-search within them — like Ctrl+F with relevance ranking, returning the exact passages along with their section and page numbers. This is how it finds where something is discussed, checks whether a term appears at all, or locates the right pages of a long document before reading them.
Reading and viewing pages
Beaver can read the actual pages of a document when a question needs continuous text — following an argument or analyzing an extended passage. When the meaning depends on layout rather than text, it looks at the page as an image instead: figures, charts, table structure, equations, and scanned PDFs with no text layer.
What Beaver can't search
There is no library-wide full-text search. Beaver cannot keyword-search the full text of every attachment in your library in one pass. Full text is only searched within papers Beaver has already identified.
In practice that means a question like "find every paper in my library that mentions social capital" gets answered in two stages: Beaver first narrows your library down using titles and abstracts, then searches the full text of those papers. Papers whose title and abstract give no hint of the topic can be missed.
Narrowing the search yourself makes this far more reliable — scope the question to a collection or tag with the @ menu (see Filtering Search), or name the papers you care about.
Beaver reads PDFs, EPUBs, web snapshots, and text or markdown files. Other formats, such as Word documents and spreadsheets, can't be read.
How it fits together
For a question like "What methods have been used to study neighborhood effects on education?", Beaver typically:
- Runs a topic search to find relevant papers
- Asks all of them the same question, or searches within them, to locate the methods discussions
- Reads the specific pages that need close analysis
- Runs a metadata search to pull in more work by the same authors
This iterative strategy lets Beaver handle questions that would otherwise take many manual searches.
You don't need to tell Beaver which search tool to use—just describe what you're looking for in natural language, and Beaver will choose the appropriate strategy.