Beaver: The AI research assistant that lives right inside Zotero

Anyone who works in academia is familiar with the problem: your Zotero bibliography keeps growing and growing – yet finding the truly relevant passage remains a laborious task. AI tools promise to provide a solution, and there is now a whole range of AI plugins and solutions, which enhance Zotero with artificial intelligence. Also MCP-based workflows offer exciting opportunities to integrate Zotero with external AI systems such as Claude or ChatGPT.

However, most of these approaches have one drawback: you have to leave Zotero and work in a different window, a different app, a different context. This is exactly where Beaver is different – an AI research assistant that lives directly in Zotero as a side panel and doesn’t interrupt your usual workflow.

What is Beaver?

Beaver was developed by Joscha Legewie , Professor of Sociology at Harvard University. Legewie is no stranger to the Zotero community: he is also the developer of ZotFile, one of the most widely used Zotero plugins of all time (4,300 stars). You can tell that Beaver has a deep understanding of Zotero – the integration is seamless.

The plugin opens as a sidebar directly within Zotero (shortcut: Cmd/Ctrl + J) and combines two core components: an advanced research agent that intelligently searches your library and synthesises results, and native Zotero integration that recognises your current reading context – right down to the page you have currently open.

Directly in the PDF: The Reading Assistant

What makes Beaver special is that it doesn’t just work at the library level, but is also integrated directly into Zotero’s PDF reader. When reading a paper, you can ask questions at any time without leaving the document:

• Highlight text passages and ask for an explanation: “What does this paragraph mean?”
• Select complex equations and get them broken down in a clear and understandable way
• Highlight graphics and illustrations and ask for an explanation
• Ask questions about the document as a whole: “Summarise this article”, “What is the main point?”, “What methodology is used?”

This becomes particularly powerful when you factor in the context of your own library. For example, you can ask questions directly from within the PDF, such as: “How does this finding compare with the other studies in my collection?” or “Which works in my library contradict this thesis?” – all without leaving the reader.

Beaver can also create annotations: if desired, it can highlight key findings or add notes that, for example, compare the results of the current paper with related studies in your own library.

Search the entire library

In addition to its role as a reading assistant, Beaver functions as a search engine for your own collection of literature. Unlike a simple keyword search, Beaver takes an ‘agent-based’ approach: it interprets research questions, combines various search strategies – from metadata queries and semantic search to full-text analysis – and refines its results iteratively.

Typical questions might include:

• “Compare all definitions of social capital in my library”
• "Which of my papers describe experimental protocols for neural plasticity?"
• "I wrote that 'studies consistently show that …' – is that correct, according to my library?"

The answers include precise source references with page numbers that can be clicked on directly – making every statement verifiable.

Discover new research

Beaver not only searches its own collection, but also provides access to over 240 million academic papers outside its own library. However, its own curated collection always remains the primary source – external results are only used to supplement this or when explicitly requested. For example, I asked Beaver what had been going on since I completed my PhD in 2015. Based on my collection, I was then presented with a truly fascinating summary of the most important topics and papers...

Organise the library and correct the metadata

As well as helping with research, Beaver also assists with the often tedious task of library maintenance. This is where the depth of the Zotero integration really comes into its own:

Automatic tagging and organisation: You can ask Beaver to automatically sort newly added papers into appropriate collections, assign tags, or categorise uncategorised entries by topic. For example: “Organise all papers added this week into appropriate collections” or “Tag my new entries as ‘to-read’”. For recurring tasks, you can define so-called actions – reusable prompts that can be accessed via keyboard shortcuts or the slash menu.

Complete missing metadata: Missing DOIs, incorrect publication years, empty abstract fields – Beaver can identify and resolve such issues. For example, it can find all entries without a DOI and look up the correct DOIs, or generate an abstract for entries where one is missing.

⚠️Important: By default, all changes require confirmation. Beaver shows you what it wants to change, and nothing is applied without your consent. If you wish, you can switch this behaviour to automatic application – you remain in control.

Supported models and costs

Beaver supports a wide range of AI models: OpenAI, Anthropic (Claude), Google Gemini, DeepSeek and other providers via custom endpoints (e.g. OpenRouter). This means maximum flexibility when choosing your preferred model.

The pricing model is fair:

• Free: 25 one-off credits to try it out. Unlimited use with your own API key.
• Plus ($10/month): 100 credits per month (carried over for one month), plus advanced features such as searching external databases, batch processing of multiple PDFs, and AI-based ranking of search results.
• Credit-Packs: 50 credits for $8 as a one-off purchase.

data protection

Beaver emphasises the protection of user data: data is not used for model training without explicit consent. Local storage options are currently being developed. The user’s own library remains the primary source of knowledge – not generic web results.

Beaver also offers an MCP server

For anyone who wants to connect Beaver to external AI tools: Beaver offers its own MCP server . This means you can also access your own Zotero library from Claude Desktop or other compatible clients – another building block for flexible research workflows.

Summary

What makes Beaver special isn’t its individual features – many AI tools can summarise texts or answer questions. The key advantage is, where Beaver works: directly within Zotero, directly within the PDF, directly within your own workflow. No switching between windows, no copying and pasting into external chatbots, no loss of context. Beaver knows which page you’re currently reading, knows your library, and can link the two together.

If you have an extensive Zotero library and are looking for an AI assistant that integrates seamlessly into your existing workflow, you should give Beaver a try.

Links:

• Beaver website
• Beaver documentation
• GitHub repository
• An overview of all AI plugins for Zotero