Zotero’s new AI roommate claims a 240M-paper memory

What it does

AI4Paper is a Zotero plugin (distributed as an XPI) that embeds a full AI research suite into the reference manager. It offers PDF translation, AI-powered literature search, reading assistance, note generation, and even automated literature reviews, all wrapped in a Zotero-native UI. The project claims to back its answers with a proprietary knowledge base of 240 million papers and citation-network data, and it supports over 20 AI providers, including local Ollama for offline use.

The interesting bit

The plugin doesn’t just chat with PDFs—it tries to own the entire workflow from discovery to manuscript. Its “AI one-click survey” crawls citation networks, filters papers across multiple rounds, and spits out a 3,000–5,000 word review with numbered references drawn from its own database. It even includes a fake-literature checker that validates reference lists against authoritative databases, a rare admission that AI hallucinations are a problem worth guarding against.

Key highlights

• Deep native integration with Zotero 7–10 (reader, item pane, notes, and sync).
• Claims a built-in RAG knowledge base of 240 million literature sources plus citation networks.
• One-click full-PDF translation preserving original layout, formulas, and charts.
• Automated literature-review generation with traceable citations and literature-matrix exports.
• Supports 20+ AI backends, from domestic Chinese models to OpenAI, Claude, and fully offline Ollama.

Caveats

• The plugin is closed-source and commercial, requiring account registration (7-day trial advertised); the repository appears to serve releases and documentation rather than source code.
• The “240 million sources” figure and citation-network coverage are vendor claims without independent verification in the README.
• Several features—such as the WeChat mini-program alerts and the proprietary AI4Paper service—are tightly coupled to the vendor’s cloud ecosystem.

Verdict

Worth a look if you live in Zotero and want an opinionated, AI-heavy research workflow, especially one optimized for Chinese-language queries and journal rankings. Skip it if you prefer open-source, self-hosted stacks or are allergic to commercial accounts and proprietary data sources.