llm-for-zotero: An AI Agent That Lives Inside Your Zotero Library

Forum topic · 小凯 · 2026-04-27

Summary

llm-for-zotero is an open-source (AGPL-3.0) Zotero 7 plugin by Yile Wang that embeds an LLM-powered research assistant directly into the Zotero reader, eliminating the need to manually upload PDFs to chatbots. Beyond basic paper Q&A, selection explanation, figure interpretation, and cross-paper comparison, it offers a Standalone Window mode, file-based notes that write Markdown with YAML frontmatter and Pandoc citations into Obsidian/Logseq vaults, and high-accuracy PDF parsing via MinerU that preserves tables, formulas, and figures. Its headline feature is Agent Mode (beta), which gives the LLM read tools (library queries, paper search, PDF page rendering, online literature search via CrossRef and Semantic Scholar) and confirmed, reversible write tools (tagging, metadata updates, collection management, item merging, imports by DOI/arXiv ID). A Skills system ships 8 built-in skills (literature review, evidence-based QA, figure analysis, etc.) and supports custom user-defined skills. The plugin supports up to four parallel models across OpenAI, Anthropic, Google Gemini, DeepSeek, and Ollama, with API Key, WebChat, and Codex App Server authentication options. At v3.7.33 with 936 GitHub stars and 972 commits, it differentiates from Aria, Zotero-GPT, and Zotero AI Butler through its full agent toolchain and deep knowledge-base integration.

Key points

llm-for-zotero is an open-source AI plugin for Zotero 7 that embeds an LLM research agent directly in the reader — no more "open PDF → drag into ChatGPT." Author Yile Wang positions it as: *"The moment you open a paper, the LLM is already ready."*

GitHub: https://github.com/yilewang/llm-for-zotero (936 stars, 53 forks, 972 commits, TypeScript 95.1%, AGPL-3.0, latest v3.7.33)
Docs: https://yilewang.github.io/llm-for-zotero/
Forum post: https://forums.zotero.org/discussion/129859/

Core features

Paper summarization with full-text auto-injection; selection explanation (up to 5 context chunks); figure interpretation (up to 10 screenshots); cross-paper comparison via / referencing across tabs
External document upload (PDF/DOCX/PPTX/TXT/Markdown)
Grounded answers with one-click jump to source text; save answers as Zotero notes with math rendering
Standalone Window Mode: full-size chat (Ctrl+Shift+L / Cmd+Shift+L), grouped conversation history, Paper chat & Library chat tabs
File-Based Notes: writes directly to Obsidian vaults / Logseq / Markdown folders with auto-filled YAML frontmatter (title, created, tags, citekey, doi, journal) and Pandoc [@citekey] citations
MinerU PDF parsing: high-accuracy extraction of tables, formulas, figures; community proxy or personal API key; local caching after first use

Agent Mode (beta)

1. Read tools (no confirmation): query_library, read_library, read_paper, search_paper, view_pdf_pages, read_attachment, search_literature_online (CrossRef, Semantic Scholar) 2. Write tools (confirmed, reversible): apply_tags, update_metadata, move_to_collection, manage_collections, manage_attachments, merge_items, trash_items, import_identifiers (DOI/ISBN/arXiv/URL), import_local_files, edit_current_note 3. System tools: file_io, run_command, zotero_script, undo_last_action (last 10 actions per session)

Skills system

simple-paper-qa

evidence-based-qa

analyze-figures

compare-papers

library-analysis

literature-review

write-note

import-cited-reference

{ZoteroDataDir}/llm-for-zotero/skills/

Multi-model architecture

Protocols: responses_api (OpenAI), openai_chat_compat, anthropic_messages (Claude), gemini_native (Google)
Built-in models include gpt-5.4 / gpt-5.4-pro, deepseek-chat / deepseek-reasoner, gemini-3-pro-preview, gemini-2.5-flash/pro, kimi-k2.5; up to 4 models in parallel with different reasoning levels
Auth modes: API Key, WebChat (browser-extension relay, no API key), Codex App Server (recommended for ChatGPT Plus users), legacy Codex Auth (being deprecated)

Comparison with alternatives

Use cases