papersgpt/papersgpt-for-zotero

PapersGPT – AI‑powered research assistant for Zotero

What it is

• A native C++ Zotero plug‑in that indexes the PDFs in your Zotero library and lets you query them with a local large language model (LLM). All processing – parsing, indexing, inference – runs on your computer, so no data leaves the device.

Key capabilities

• Massive‑scale indexing – 10 000 PDFs (≈42 GB, 170 k pages) can be indexed in ~7 minutes; the resulting index is ~900 MB.
• Instant retrieval – typical query latency is ~20 ms, even on the largest library.
• Cross‑document synthesis – ask questions that span many papers (e.g., “compare sample sizes and p‑values of all cardiology trials”) and get structured answers with clickable citations that jump to the source PDF.
• Zero‑cloud, zero‑hallucination – the engine works entirely offline; every answer is backed by a citation from your library.
• Local LLM support – choose from built‑in open‑source models (Gemma 4 12B, Qwen3.5 4B), an OpenAI‑compatible private endpoint, or any model running via Ollama.
• AutoPilot – define a research goal and let the plug‑in batch‑process thousands of papers overnight, automatically creating Zotero notes.
• MCP/Skill integration – expose the indexed library as a “skill” for agentic platforms (Claude, Cursor, DocsAgent, etc.) so other tools can query your Zotero data.

Typical workflow

1. Install the papersgpt‑vX.x.x.xpi file from the GitHub release page.
2. Open a PDF, a selection of PDFs, or the whole library and click the PapersGPT icon (or use Ctrl+Enter / Cmd+Enter).
3. Pick a local LLM (or configure a private endpoint) and type a question or use a built‑in prompt (summary, literature‑review, etc.).
4. The plug‑in returns a concise answer with inline citations; click a citation to jump to the exact page in Zotero.
5. Save insights as Zotero notes or close the chat window.

Who benefits

• Academic researchers who need to synthesize large corpora of papers without uploading them to a cloud service.
• Legal and finance professionals who must search thousands of case files or filings while keeping data on‑premises.
• Developers of AI agents who want a fast, local knowledge base that can be called from Claude, OpenAI, or other agent frameworks.

Performance snapshot

Library size	Raw size	Index time	Avg. query	RAM used	Index size
1 000 PDFs	4.2 GB	51.5 s	13 ms	353 MB	100 MB
10 000 PDFs	42 GB	421 s (7 min)	19.5 ms	2.21 GB	901 MB

Getting started

• Download the latest .xpi from the releases page.
• Follow the quick‑start guide on papersgpt.com to install the plug‑in in Zotero.
• Configure a local LLM (e.g., download Gemma 4 12B via the built‑in selector) and start chatting.

Why it matters PapersGPT brings industrial‑scale, low‑latency semantic search to a desktop reference manager while guaranteeing 100 % data sovereignty. It lets users move from simple keyword search to true multi‑document reasoning without relying on external APIs.