All posts
product2026/11/12

BibGenie vs llm-for-zotero: Which Zotero AI Plugin Should You Choose?

Compare BibGenie and llm-for-zotero as Zotero AI plugins. See how an open-source Zotero agent with CLI and WebChat options stacks up against a zero-config plugin with semantic search, metadata review, local models, and Zotero MCP.

Both BibGenie and llm-for-zotero answer questions inside Zotero with clickable citations back to your source passages, support PDF reader screenshot queries for figures and tables, and include a built-in Zotero MCP server. The real difference is not what they promise — it is what you have to configure to get there, and who runs the underlying machinery.

Quick Verdict

  • Choose llm-for-zotero if you want maximum freedom and enjoy configuring tools. The plugin is currently free to install and open source (AGPL-3.0), its Agent Mode offers deep library automation with local full-text indexing, and it provides multiple ways to run without an API key (WebChat, Codex App Server, Claude Code).
  • Choose BibGenie if you want a zero-config llm-for-zotero alternative to read papers and organize your library immediately. No API key, no CLI, no Node.js, and no browser extension: install it, and hosted models work out of the box alongside an automatic title/abstract semantic index and a 5-source metadata review workflow with diff previews.

Comparison at a glance

DimensionBibGeniellm-for-zotero
Citation traceabilityClick a citation to jump to the exact PDF passage, with the text block highlightedClick a citation to jump back to the source passage from the reader
Price (plugin + model)Free 200 credits/month; Pro $9.99/month (10,000 credits); Lifetime $199 early-bird one-time (retiring after v1.0)Plugin currently free to install; you pay your model provider, reuse a ChatGPT Plus / Claude Code subscription, or use WebChat with no key
Free tier200 hosted credits/month, no API key neededPlugin currently free to install; model cost is yours
Models & local modelsHosted models, BYOK, local models via Ollama / LM StudioOpenAI, Anthropic, Gemini, DeepSeek, Grok, Qwen, Kimi, OpenAI-compatible endpoints, local models, WebChat, Codex App Server, Claude Code
Hosted modelBuilt-in hosted models — no key, subscription, or local setup neededNone — every backend is your own key, local model, or existing ChatGPT Plus / Claude Code / web-chat account
Search & library indexAutomatic local semantic index over titles, abstracts, and metadata — zero setupLocal full-text indexing & optional vector search (since v3.9.10; indexing fixes ongoing in v3.9.17), plus conversation-context tracking
Agent depth & reader UXChat (including PDF screenshot queries) plus library operations with diff preview; built-in MCP serverAgent Mode (beta): search, tag, metadata, import, notes, collections; 8 built-in + custom skills; Task progress panel; Inline Ask on PDF selections (v3.9.17); file & command tools
Zotero MCP serverBuilt-in Zotero MCP serverBuilt-in MCP server
Privacy & data flowChat history and index stay local by default; hosted route uses BibGenie's service; local chat still depends on separate embedding & external tool settingsAPI keys go to your provider; local models stay local; WebChat routes through chatgpt.com / chat.deepseek.com / gemini.google.com; cloud MinerU sends PDFs to mineru.net
Library organizationBuilt-in library management; every change is previewed as a diff card before executionAgent tools apply tags, update metadata, manage collections, merge duplicates, and trash items with approval
Metadata reviewDedicated review workflow cross-checking fields against arXiv, Semantic Scholar, OpenAlex, DBLP, and OpenReview with diff previewBuilt-in Audit Library (health report note) and Complete Metadata actions in Agent Mode, fetching fields via CrossRef & Semantic Scholar
Note-taking & syncSaves answers as new standalone Zotero notesExports Markdown notes (with equations & screenshots); companion ecosystem for Obsidian / Logseq workflows
Search beyond libraryBuilt-in OpenAlex discovery for papers outside your libraryAgent skills (literature_search, literature_review) search OpenAlex, arXiv, and Europe PMC; /discover_related command
Zotero compatibilityZotero 7–10Zotero 7–10, 11 beta
Maintenancev0.9.6 (Nov 10, 2026)v3.9.17 (Nov 10, 2026); 121 releases since Jan 2026; 3,549 GitHub stars
Ease of setupInstall and go; hosted models need no keyInstall .xpi, configure provider & test — or set up WebChat (browser extension + visible tab), Codex (npm), or Claude Code (Node.js 18+)
Openness / auditabilityClosed sourceAGPL-3.0 open source; Atlas code rating C on v3.9.17 (static code check, not yet tested in live Zotero)

Citation traceability: both anchor answers to the source

This is the dimension that matters most for research, and here the two plugins are genuinely comparable. Both anchor answers in the source text, support asking multimodal questions via PDF reader screenshots, and let you click a citation to jump back to the passage it came from.

BibGenie highlights the cited text block on the PDF page (and supports EPUBs, web snapshots, and native annotations); llm-for-zotero jumps directly to the source passage from the reader and highlights cited sentences via its Task progress view ("Show in answer"). Citation Styler's overview of Zotero AI plugins describes llm-for-zotero's answers the same way: "always anchored in the source text — with a single click, you can jump back to the relevant passage." BibGenie citation tracing

One honest caveat applies to both: a real source does not guarantee a correct interpretation. A citation pointing to the right paragraph can still be synthesized into an inaccurate claim. Traceability makes checking fast; it does not replace checking.

This one: both deliver reliable source-anchored citations, differing mainly in how highlights and jump links are displayed.

Real total price: free to install vs managed

On plugin licensing, llm-for-zotero is currently free to install under an AGPL-3.0 open-source licence. What you pay is whatever your model costs. That can be close to zero — local models via Ollama, WebChat with no key, or a ChatGPT Plus / Claude Code subscription you already pay for — or an API bill that scales with usage.

BibGenie is a managed service, so the arithmetic is different. The Free plan includes 200 hosted credits a month with no key required. Pro is $9.99/month for 10,000 credits, and a one-time Lifetime Pro early-bird plan ($199) is currently available and will retire after the v1.0 release (existing Lifetime buyers keep their benefits). Bringing your own key or running local models is free of platform model charges on either plan; credits only pay for the hosted model route and supplementary external tools. BibGenie pricing

So the price question is really: do you want to pay with money, or with setup and maintenance? llm-for-zotero is cheaper in money; BibGenie is cheaper in time.

This one: both let you bring your own key or local models; choose llm-for-zotero if you want an open-source plugin with web or CLI subscription reuse, or BibGenie if you want built-in hosted credits out of the box.

Models and Zotero local models: both support major providers and local models

Both plugins support bring-your-own API keys from major providers (OpenAI, Anthropic, Gemini, and OpenAI-compatible endpoints) as well as running local models via Ollama and LM Studio.

Beyond that shared foundation, each adds a different convenience layer. llm-for-zotero supports additional Chinese providers (DeepSeek, Grok, Qwen, Kimi) and three no-key bridge paths: WebChat (via a browser extension syncing with chatgpt.com, chat.deepseek.com, or gemini.google.com), the Codex App Server (reusing a ChatGPT Plus subscription), and Claude Code (bridging your local Claude Code login). llm-for-zotero on GitHub

BibGenie focuses on zero-friction readiness: alongside BYOK and local models via Ollama or LM Studio (with independent configuration for chat and embedding models), it is the only one of the two with built-in hosted models that work out of the box with no API key or local hardware required. BibGenie custom and local models

This one: both support major API keys and local models (Ollama / LM Studio); llm-for-zotero adds extra web and CLI bridges, while BibGenie adds zero-setup hosted models.

Agent depth and Zotero MCP: comparable core workflows, different focus

On the core research loop — library retrieval, paper reading, citation-backed evidence, and library actions — both plugins offer comparable capabilities, with different workflow philosophies.

llm-for-zotero's Agent Mode (beta, off by default) operates across your library to read, search, tag, edit metadata, import papers, and write or edit notes. It ships with 8 built-in skills and custom skill authoring. For multi-step workflows (replacing its earlier standalone \plan mode removed in v3.9.10), its Task progress panel shows which paper sections the agent read, why items were kept or skipped, and external sources used. In v3.9.17, it also added Inline Ask right next to PDF text selections, alongside Tavily web search, MinerU parsing, and Markdown note exports for Obsidian or Logseq. llm-for-zotero documentation

BibGenie packages retrieval, reading, evidence tracing, and library management into a cohesive zero-config workflow: grounded chat with highlighted PDF citations (including PDF reader screenshot analysis), OpenAlex discovery, systematic metadata review across 5 academic databases, and library changes previewed as diff cards before anything is written — without requiring skill scripts or external tooling.

Both plugins include a built-in Zotero MCP server, allowing external AI clients such as Claude Desktop or Cursor to connect to your library. And both require confirmation before writing changes to Zotero — the difference is that BibGenie presents a visual diff card preview upfront, whereas llm-for-zotero relies on agent-step approvals. BibGenie MCP documentation

This one: comparable on core retrieval, reading, citation-backed answers, and Zotero MCP; llm-for-zotero leans toward custom skills and CLI extensibility, while BibGenie focuses on zero-config workflows and diff-previewed edits.

Privacy and data flow: depends on the route you pick

Neither plugin has a single privacy answer, because both change behaviour depending on the backend you choose.

With BibGenie, chat history and the library semantic index stay on your machine by default. The hosted model route sends prompts to BibGenie's model service; bring-your-own-key requests connect directly from Zotero to your provider. Note that even when you configure a local chat model (Ollama / LM Studio), your embedding model and external tools (such as web search or OpenAlex discovery) still follow their own separate endpoint settings — to keep everything strictly on-device, configure both chat and embedding models locally and avoid online tools. BibGenie custom and local models

With llm-for-zotero, API-key requests go directly to your provider, and local models stay local. However, two of its most popular convenience routes send data elsewhere: WebChat routes questions through third-party web apps (chatgpt.com, deepseek.com, gemini.google.com), and cloud MinerU sends newly parsed PDFs to mineru.net (though a self-hosted local MinerU server is supported as an alternative). In practice, choosing a "free, no-key" path often means trading away local data isolation.

This one: on local/BYOK routes, both stay local-first when every component is configured locally, while BibGenie avoids third-party web-chat proxies.

Search and indexing: automatic title/abstract index vs full-text & vector options

Both plugins now support local indexing, but they make different trade-offs in coverage, setup effort, and entry points.

BibGenie automatically builds and maintains a local semantic index primarily over your library's titles, abstracts, and metadata (it is not a full-text PDF vector index). There is nothing to configure: out of the box, you can search across your entire collection by meaning and let the assistant retrieve relevant papers before answering, even when you forgot the exact title or author. BibGenie semantic index

llm-for-zotero introduced local full-text indexing and optional vector search in v3.9.10 (with indexing behavior continuing to receive fixes in v3.9.17), alongside its Task progress view that tracks which paper sections were read. Compared to BibGenie's title-and-abstract index, llm-for-zotero's full-text indexing reaches deeper into PDF body text once configured and indexed — though it involves more setup and is still actively stabilizing.

This one: a split decision — BibGenie for zero-config meaning search over titles and abstracts; llm-for-zotero if you want local full-text PDF indexing and are willing to configure it.

Organizing your library and checking metadata: structured diff review vs agent audit skills

Both plugins can organize a library and audit metadata, though their workflows differ in sources, completion mechanics, and previewing.

For metadata verification, BibGenie provides a dedicated review workflow that cross-checks fields against arXiv, Semantic Scholar, OpenAlex, DBLP, and OpenReview. Proposed corrections are rendered as a field-level diff card so you can inspect and approve exact changes before anything is written. BibGenie library management

llm-for-zotero includes built-in Audit Library and Complete Metadata actions in its Agent Mode. Audit Library scans your library or collection for structural gaps (incomplete metadata, missing PDFs, absent tags) and can save the health report as a Zotero note; Complete Metadata fetches canonical bibliographic fields (such as via CrossRef and Semantic Scholar) to fill missing records with step approval.

For general organization and discovery, BibGenie provides built-in diff-previewed management and a dedicated OpenAlex discovery UI; llm-for-zotero applies tags, updates collections, merges duplicates, and searches OpenAlex, arXiv, and Europe PMC via agent skills (literature_search, literature_review) and /discover_related.

This one: BibGenie for 5-source field cross-checking with visual diff cards; llm-for-zotero for library-wide structural health audits and automated agent completion.

Ease of setup: BibGenie is install-and-go

BibGenie installs from the .xpi plugin file and works immediately: hosted models require no API key, no external provider account, and no command line. BYOK and local models are optional upgrades.

llm-for-zotero asks more of you up front. The standard path requires configuring a provider base URL, API key, and model name. The no-key paths each add moving parts: WebChat requires a browser extension and an open Chromium tab; Codex requires npm install -g @openai/codex and CLI login; Claude Code requires Node.js 18+. Enabling optional vector search and local MinerU adds further configuration. One quirk worth noting: several forks of llm-for-zotero share the same add-on ID, so Zotero can only have one installed at a time.

This one: BibGenie, for most non-developer researchers.

Maintenance: both are actively developed

llm-for-zotero is one of the fastest-moving Zotero plugins available: launched in January 2026, it reached v3.9.17 on November 10, 2026 (its 121st release), accumulating 3,549 GitHub stars and nearly 900,000 recorded downloads. That velocity brings rapid features alongside frequent changes to keep up with.

BibGenie moves at a steadier pace: v0.9.6, also released on November 10, 2026, focused on usage tracking and sync improvements. Its public releases repository carries 18 stars, while core development happens in a private repository.

This one: llm-for-zotero on velocity; BibGenie if you prefer fewer moving parts.

Openness and auditability: the honest trade

Here the two products are mirror images.

llm-for-zotero is fully open source under AGPL-3.0. Anyone can inspect the code, and the Zotero Plugins Atlas does: its latest audit of v3.9.17 (checked November 12, 2026) assigns llm-for-zotero a grade C ("Not recommended"), specifically flagging that it installs and runs npm or PyPI packages without a fixed version, alongside the agent's ability to execute system commands. Importantly, the Atlas page explicitly notes that this rating is based on a static code check and has not yet been tested in an actual Zotero environment — it is a supply-chain and surface-area audit finding, not a report of user harm.

BibGenie is closed source. Its code cannot be audited externally the same way, which is a genuine transparency limitation. Instead, BibGenie offers a narrower operational trust surface: local-first storage by default, no prompts, responses, or keys uploaded in optional usage sync, and diff-previewed library writes without executing unpinned external packages.

This one: llm-for-zotero if you require auditable open-source code; BibGenie if you want a smaller execution surface without unpinned package installs.

Strengths and limitations

llm-for-zotero

Strengths

  • Currently free to install and open source (AGPL-3.0)
  • Highly extensible agent workflow: skills, Task progress tracking, Audit Library / Complete Metadata, Inline Ask, file and command tools
  • Widest no-key backend support: WebChat, Codex App Server, Claude Code, local models
  • Local full-text indexing and optional vector search (since v3.9.10)
  • MinerU high-fidelity PDF parsing, clean Markdown export (equations & screenshots), and Obsidian / Logseq companion ecosystem
  • Massive, fast-moving community (3,549 GitHub stars, 121 releases)

Limitations

  • Complex setup depending on backend (extensions, Node.js, CLI logins, optional vector/MinerU tuning)
  • Agent Mode remains in beta, and full-text indexing continues to receive fixes (v3.9.17)
  • Zotero Plugins Atlas grade C on v3.9.17 (static code check, untested in live Zotero) flagging unpinned npm/PyPI package installs and command execution
  • Cloud MinerU sends newly parsed PDFs to mineru.net; WebChat routes data via web chat portals
  • Forks share the same add-on ID, preventing side-by-side installation

BibGenie

Strengths

  • Install-and-go hosted models included (200 free credits/month, no API key or CLI needed)
  • Automatic local semantic index over titles, abstracts, and metadata with zero setup
  • Traceable citations back to exact highlighted PDF passages, EPUBs, snapshots, and native annotations (plus PDF reader screenshot queries)
  • Preview-before-write diff cards for library edits
  • Built-in OpenAlex discovery and 5-source metadata review workflow
  • Narrower execution surface; BYOK and local models free of platform model charges

Limitations

  • Closed source; cannot be independently audited like AGPL-3.0 projects
  • Semantic index covers titles, abstracts, and metadata rather than full-text PDF embeddings
  • Hosted model route requires credits beyond the 200/month free tier
  • Shallower agent surface: no custom skills framework, shell command tools, or Inline Ask popups
  • Quick Prompts trigger from chat only; saving notes creates new notes rather than appending
  • Newer plugin with fewer independent third-party reviews so far

Who should choose whom

  • Choose llm-for-zotero if you want maximum freedom, enjoy tinkering with developer tools, want custom skills with full-text indexing and Audit Library automation, value AGPL-3.0 open-source code, or want to reuse an existing ChatGPT Plus or Claude Code subscription.
  • Choose BibGenie if you are looking for a zero-config llm-for-zotero alternative — install one plugin, start reading with AI immediately, search across titles and abstracts automatically via semantic indexing, and verify metadata against 5 academic databases with diff cards without touching Node.js or CLI logins.
  • Consider running both if you want llm-for-zotero's custom skills or CLI bridges for specialized automation and BibGenie's zero-config chat, automatic semantic index, and 5-source metadata review for everyday research.

Frequently Asked Questions