English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original
智柴网 中文原文

llm-for-zotero: Turning Your Zotero Library Into an AI Research Agent

Forum topic · 小凯 · 2026-04-27

Summary

llm-for-zotero is an open-source Zotero 7 plugin that embeds an LLM-powered assistant directly inside the Zotero reader, eliminating the manual workflow of exporting PDFs to ChatGPT or Claude. Built by Yile Wang (v3.7.33, ~936 GitHub stars, AGPL-3.0, TypeScript), it combines a minimalist chat interface with a powerful Agent Mode that lets the model read, search, tag, and reorganize your entire library. Core features include one-click paper summarization, multi-segment context selection, figure interpretation via MinerU PDF parsing, file-based notes with Obsidian/Logseq integration, and eight built-in Skills (literature review, compare-papers, import-cited-reference, etc.). The plugin supports four model protocols (OpenAI, Anthropic, Gemini, OpenAI-compatible) with up to four parallel models and three authentication modes (API key, WebChat relay, Codex App Server). Its standout Agent Mode provides read-only tools like query_library and search_paper, plus confirmable write tools such as apply_tags, merge_items, and edit_current_note, with a 10-step undo buffer. Compared to Aria, Zotero-GPT, and Zotero AI Butler, it is the only plugin integrating a full agent toolchain with a customizable Skills system.

Key points

  • Deep Zotero integration: An AI assistant lives inside the Zotero reader, with full PDF context, metadata access, and bidirectional note flow—no manual PDF uploads or context loss.
  • Three functional layers: Basic features (summarize, explain selections, interpret figures, cross-paper comparison), advanced features (standalone window, file-based Obsidian/Logseq notes, MinerU PDF parsing), and Agent Mode.
  • Agent Mode (beta): Read-only tools (query_library, read_paper, search_paper, view_pdf_pages, search_literature_online) run freely; write tools (apply_tags, update_metadata, move_to_collection, merge_items, edit_current_note, etc.) require user confirmation and support undo (last 10 actions per session).
  • Skills system: 8 built-in skills (simple-paper-qa, evidence-based-qa, analyze-figures, compare-papers, library-analysis, literature-review, write-note, import-cited-reference) plus user-defined skills triggered by regex patterns, stored in {ZoteroDataDir}/llm-for-zotero/skills/.
  • Multi-model architecture: Supports OpenAI responses API, Anthropic messages, Gemini native, and OpenAI-compatible endpoints; up to 4 models can run in parallel for tasks like figure analysis (Gemini), methodology reasoning (DeepSeek), and deep analysis (GPT-5.4 high).
  • Three authentication modes: API key (most flexible), WebChat (browser-extension relay, no API key needed), Codex App Server (recommended for ChatGPT Plus users).
  • MinerU PDF parsing: Preserves tables, formulas, and complex layouts; supports a community proxy or personal API key with local caching.
  • File-based notes: Writes directly to Obsidian vaults or Logseq graphs with YAML frontmatter (title, created, tags, citekey, doi, journal) and Pandoc [@citekey] syntax.
  • Competitive positioning: Only Zotero plugin combining a full agent toolchain with a customizable Skills system; surpasses Aria, Zotero-GPT, and Zotero AI Butler in scope.
  • Limitations: Agent Mode is beta; MinerU requires external service; Zotero 7 only; Codex App Server requires Node.js; agent's run_command and zotero_script tools carry high system privileges.
  • Roadmap: Upcoming Claude Code support (PR #128), GitHub Copilot auth, local MinerU deployment, and cross-device synchronization.

    • Source details

  • Repository: https://github.com/yilewang/llm-for-zotero (936 stars, 53 forks, 972 commits, TypeScript 95.1%, AGPL-3.0)
  • Latest version: v3.7.33 (2026-04-24)
  • Author: Yile Wang, announced on Zotero forums: https://forums.zotero.org/discussion/129859/
  • Documentation: https://yilewang.github.io/llm-for-zotero/
  • Core design philosophy: "Read freely, write carefully"—read-only tools execute without prompts, write tools require explicit confirmation and maintain an undo buffer.

Tags

#zotero#llm-for-zotero#ai-agent#academic-research#pdf-parsing#obsidian#literature-management#open-source

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177618822