LLM-for-Zotero Documentation

LLM-for-Zotero

A Zotero plugin that brings large language models and agent workflows directly into the PDF reader so you can summarize papers, explain selected text, interpret figures, compare related work, and save useful outputs back into your notes.

Zotero 7 and 8 OpenAI, Gemini, DeepSeek, OpenAI-compatible APIs Codex auth supported Notes, screenshots, and file uploads Agent mode with approvals
GitHub Download .xpi

Quick Start

  1. Install the latest .xpi from the release page.
  2. Open Zotero preferences and configure your model provider.
  3. Open any PDF and launch the LLM sidebar from the right toolbar.
  4. Ask a question such as Summarize this paper in five bullets.
LLM for Zotero interface

Overview

LLM for Zotero is designed for paper reading inside Zotero rather than in a separate browser tool. The plugin keeps your paper, context, screenshots, and notes in one workflow.

Read in place

Open a paper in Zotero and chat with the current PDF without leaving the reader.

Ground answers in context

Use selected text, screenshots, or extra files to give the model exactly the evidence it needs.

Work across papers

Open multiple tabs and cite another paper with / when you want comparison or synthesis.

Keep results in Zotero

Save answers, selected snippets, screenshots, and conversation history directly into notes.

Install and configure

You only need a Zotero install and one working model provider connection.

Install the plugin

Download the latest .xpi from GitHub Releases, then in Zotero open Tools -> Add-ons and choose Install Add-on From File.

Open the settings panel

Restart Zotero, open Preferences, and switch to the llm-for-zotero tab.

Add a model connection

Pick your provider, then enter the API URL, secret key, and model name. You can set up multiple providers and models for different tasks.

Test before using

Use Test Connection to verify the configuration before starting your first paper chat.

Good first question

What is the main conclusion of this paper, and what evidence supports it?

Supported extra files

PDF, DOCX, PPTX, TXT, and Markdown files can be added as extra context.

Daily workflow

A typical reading session fits into four moves.

1. Start broad

Ask for a summary, the research question, or the main takeaway to decide whether the paper is worth deeper reading.

2. Narrow with context

Select a paragraph, add a figure screenshot, or upload a related file when a specific claim needs more explanation.

3. Compare or extend

Use / to cite another open paper and ask how the methods, datasets, or conclusions differ.

4. Save what matters

Export a strong answer into Zotero notes so your reasoning and supporting context stay attached to the paper.

Main functions

These are the core tasks the plugin is built to handle.

Summarize a paper quickly

The first turn loads the full paper context so the model can produce a summary grounded in the document you are reading.

Summarize this paper in five bullets: problem, method, data, main result, and limitation.

Summarizing a paper in LLM for Zotero

Explain selected text

Select a dense paragraph or term in the PDF and ask the model to explain it. The selected text becomes extra context on top of the paper itself.

Explain this selected paragraph in plain English and define the technical terms.

Explaining selected text in LLM for Zotero

Interpret figures and screenshots

Capture a figure or visual region from the paper and ask for an interpretation. This is useful when the key claim is hidden in a chart or diagram.

What does this figure show, and how does it support the paper's argument?

Interpreting a figure in LLM for Zotero

Compare multiple papers

When multiple papers are open in Zotero, type / to cite another paper as context and ask for a structured comparison.

Compare this paper with /the other tab on objective, method, dataset, and weaknesses.

Comparing papers in LLM for Zotero

Add extra documents

You can upload local files that are not already in Zotero, including PDF, DOCX, PPTX, TXT, and Markdown.

Read this uploaded document and tell me how it changes the interpretation of the current paper.

Uploading extra files in LLM for Zotero

Save answers and conversation history

Strong answers can be exported directly into Zotero notes, and conversation history is stored per paper so you can resume later.

Save this answer to my note with headings and preserve the equations.

Saving an answer to a note
Saving a conversation in Zotero

Custom shortcuts

Create quick actions for repeated workflows like methodology summaries, limitations, or future-work extraction.

Agent Mode

Switch to agent mode when you want the plugin to inspect your library, plan multi-step work, and prepare changes for review.

Agent mode

Chat mode is best for direct paper questions. Agent mode is for workflows where the assistant needs to inspect Zotero state, choose tools, and complete several steps in sequence.

How to enable it

  1. Open Preferences and go to the llm-for-zotero tab.
  2. Check Enable Agent Mode (Beta).
  3. Open a PDF and click the Agent mode toggle in the context bar.

The toggle stays hidden until the beta setting is enabled.

When to use it

  • Use chat mode for fast answers about the current paper.
  • Use agent mode for library search, note inspection, metadata cleanup, tagging, and collection work.
  • Run it from the current paper or from open chat when the task spans multiple papers and files.

What the agent can inspect

It can search your library, browse collections, list unfiled or untagged papers, read notes and annotations, compare papers, search related work online, and inspect the current reader view or specific PDF pages.

Safe write actions

When a task would modify Zotero data, the agent prepares a review card first. You approve note saves, metadata edits, tag changes, new collections, and collection moves before anything is applied.

Visible activity trace

Each agent turn shows an activity log in the chat so you can see tool calls, status updates, and expandable reasoning summaries instead of getting a black-box answer.

Undo support

The agent keeps an undo stack for recent write actions in the current conversation, so you can ask it to undo the last agent-applied change if a batch edit is not what you wanted.

Good agent prompts

  • Find papers in my Zotero library related to this article and group them by method.
  • Audit this paper's metadata, explain what is wrong, and prepare a clean fix.
  • Read my notes and annotations for this paper, then draft a structured summary.
  • List untagged papers in this library and suggest tags I can review.

Codex auth

If you have a ChatGPT Plus subscription, you can reuse that login with Codex auth instead of supplying an API key.

Setup steps

  1. Install the Codex CLI.
  2. Run codex login in your terminal.
  3. In Zotero settings, set Auth Mode to codex auth.
  4. Set API URL to https://chatgpt.com/backend-api/codex/responses.
  5. Choose a Codex model such as gpt-5.4.
  6. Click Test Connection.

What this enables

  • Use Codex models without an API key.
  • Ground answers in local PDF text and screenshots.
  • Reuse credentials stored in ~/.codex/auth.json.
  • Keep the same Zotero workflow while lowering API key management overhead.
Codex auth setup in LLM for Zotero

Starter prompts

A few prompts that work well on day one.

Summarize this paper for me in five bullets and tell me whether it is worth a careful read.
Explain this selected text like I am new to this subfield, but keep the scientific meaning precise.
What assumptions does this method make, and where in the paper are they justified?
Compare this paper with /the other open paper on data, method, evaluation, and limitations.
Read this uploaded document and tell me which parts are relevant to the paper I am reading now.
Turn the best part of this answer into a Zotero note with a short title and bullet-point takeaways.