Please see the latest release notes. The plugin name is now changed to llm-for-zotero.
Recent Update: 🤖 LLM-for-Zotero now becomes your own Agent in your Zotero app (BETA)! Please check the Agent Mode section for more details.
LLM-for-Zotero now becomes your own Zotero Agent! It is currently in beta and is disabled by default. To enable it, please go to the settings and check the Enable Agent Mode checkbox.
When you enable the agent mode, you will see the Agent (beta) toggle in the context bar. Click it to switch to agent mode.
However, the agent mode is still in beta and is disabled by default. The functionality is still limited and is not fully tested. I recommend you to use either Codex, Gemini or Anthropic models to get the best performance. If you have any feedback or suggestions, please feel free to open an issue on GitHub!
Previous Update: ChatGPT Plus subscription users can use their Codex quote to access Codex models (e.g. gpt-5.4) without an API key. Please see the Usage Guide for more details.
llm-for-zotero is a powerful plugin for Zotero that integrates Large Language Models (LLMs) directly into the Zotero PDF reader. Unlike other tools that require you to upload your pdfs to a portal, this plugin is designed to conveniently access LLMs without the need to leave Zotero. It quietly sits in the panel of the Zotero reader, like your standby research assistant, ready to help you with any questions you have when reading a paper.
It is always the first question that comes to mind when you open a new paper. With this plugin, you can get a concise summary of the paper in seconds, without having to read through the entire text. You can always find the relevant information you need, and quickly decide whether this paper is worth your time. The summary is generated based on the full content of the paper currently open in your Zotero reader, so you can be sure that the information is accurate and relevant. You can also customize the summary prompt to fit your specific research needs, such as focusing on the methodology, results, or implications of the paper.
If you come across a complex paragraph or a technical term that you don't understand, simply select the text and ask the model to explain it.
You can select up to 5 contexts from both model's answer and the paper content to provide more context for the model to generate a more accurate and detailed explanation.
In the plugin, you can enable the pop-up option to conveniently add text to the chat.
If you don't like it, that's totally fine. You can always disable it in the settings! I really think it is important to give you the choice.
When you start your chat with the model, the full context of this paper is loaded into the model on the first turn. Follow-up questions switch to focused retrieval from the same paper, so the selected text becomes an additional layer of context that helps the model give a more accurate and detailed explanation without resending the whole paper each time.
In our research, understanding the figures equivalent to understanding the paper. With this plugin, you can take a screenshot of any figure in the paper and ask the model to interpret it for you. It supports up to 10 screenshots at a time.
In this plugin, you can open multiple papers in different tabs and ask the model to read and compare the papers for you. You can simply type / to cite the other paper as additional context.
Not limited by the papers in your Zotero library, you can also upload any document from your local drive to the model as additional context. The supported file types include PDF, DOCX, PPTX, TXT, and markdown files. This feature is developed by coder contributor @jianghao-zhang. Kudos to him!
This plugin supports seamless integration with your note-taking workflow. You can easily save an answer, or selected text generated by the model into your Zotero notes with just one click.
The local conversation history is automatically saved and associated with the paper you are reading. You can also export the whole conversation into your note with markdown format. I spent a plenty amount effort to make sure the mathematical equations rendering is correct in the exported markdown, so complex knowledge will be presented in the most clear way in your note.
The export to note function also supports saving the selected text and screenshots into your note with markdown format. So, you never lose context.
You can customize the quick-action presets to fit your specific research workflow.
You can set up multiple provider connections, and each provider can include multiple LLM models to help you handle different types of tasks. For instance,
- the multimodal model for helping you to interpret the Figure;
- text-based model for helping you understand text.
Different models can also be used for the same task, and you can cross check their answers to get a more comprehensive understanding of the paper.
If you want more customization, you can also set up different reasoning levels for the same model in the conversation panel, such as "default", "low", "medium", "high" and "xhigh" for gpt-5.4, "medium", "high" and "xhigh" for gpt-5.4-pro, "low" and "high" for gemini-3-pro-preview, "medium" for gemini-2.5-flash. You can always check the connections by clicking the "Test Connection" button in the settings.
If you are a pro player, you can also change some hyperparameters of the model, such as temperature, max_tokens_output, etc. to get more creative or more deterministic answers.
The plugin will automatically check for updates when you open Zotero. If I release an update, you will be able to update the plugin with just one click, without having to go through the installation process again. This way, you can always enjoy the latest features and improvements without any hassle.
The public plugin name is now llm-for-zotero.
To keep auto-update seamless for existing installs, the internal add-on ID remains unchanged from earlier releases.
User settings and local chat history are migrated automatically on startup.
Download the latest .xpi release from the Releases Page.
Open Zotero and go to Tools -> Add-ons.
Click the gear icon and select Install Add-on From File
Select the .xpi file and restart Zotero to complete the installation.
Open Preferences and navigate to the llm-for-zotero tab.
Select your Provider (e.g., OpenAI, Gemini, Deepseek).
Paste your API Base URL, secret key and model name.
This plugin also supports codex auth mode (ChatGPT/Codex login reuse) for Codex models:
- Run
codex loginon your machine first. - In plugin settings, set provider Auth Mode to
codex auth. - Use API URL
https://chatgpt.com/backend-api/codex/responsesand a Codex model (e.g.gpt-5.4).
codex auth v1 notes:
- The plugin reads local credentials from
~/.codex/auth.json(or$CODEX_HOME/auth.json). - If request returns 401, the plugin attempts token refresh automatically.
- Embeddings are not supported in codex auth mode yet.
- Local PDF/reference text grounding and screenshot/image inputs are supported in codex auth mode.
- Only Responses
/filesupload +file_idattachment flow is not supported in codex auth mode yet.
I will give some popular model as example:
| API url | Model Name | Reasoning Level | Notes |
|---|---|---|---|
| https://api.openai.com/v1/responses | gpt-5.4 | default, low, medium, high, xhigh | PDF uploads supported |
| https://api.openai.com/v1/responses | gpt-5.4-pro | medium, high, xhigh | PDF uploads supported |
| https://api.deepseek.com/v1 | deepseek-chat | default | |
| https://api.deepseek.com/v1 | deepseek-reasoner | default | |
| https://generativelanguage.googleapis.com | gemini-3-pro-preview | low, high | |
| https://generativelanguage.googleapis.com | gemini-2.5-flash | medium | |
| https://generativelanguage.googleapis.com | gemini-2.5-pro | default, low, high | |
| https://api.moonshot.ai/v1 | kimi-k2.5 | default |
Now, the plugin natively supports different provider protocols, such as responses_api, openai_chat_compat, anthropic_messages, gemini_native, etc. You can choose the one that suits your needs.
You can always check the connections by clicking the "Test Connection" button.
When you enable the agent mode in the preferences settings, you will see the Agent (beta) toggle in the context bar. Click it to switch to agent mode.
Currently, the agent mode supports the following tools:
search_related_papers_online: Search for related or similar papers online.read_paper_evidence: Read the paper evidence.read_paper_excerpt: Read the paper excerpt.prepare_pdf_pages_for_model: Prepare the PDF pages for the model.write_note: Write a note.apply_tags: Apply tags to the paper.- ... And more!
I only show some of the tools in the demos below. You can always check the tools by asking the agent what it can do.
More tools will be supported in the future. And I also welcome you to contribute your own tools to the agent!!
This is one the first step for this agent mode. In the future, it should be a versatile agent mastering all the tasks in your Zotero library.
To chat with a paper, open any PDF in the Zotero reader.
Open the LLM Assistant sidebar (click the distinct icon in the right-hand toolbar).
Type a question in the chat box, such as "What is the main conclusion of this paper?"
We know token consumption is a big concern for many users. If you have a ChatGPT Plus subscription, you can use Codex auth mode to access Codex models (e.g. gpt-5.4) without an API key. The plugin reuses your ChatGPT login via the Codex CLI. This is a great way to save your tokens. Special thanks to @jianghao-zhang for their valuable contributions to this project.
Step-by-step setup (including new computers):
-
Install the Codex CLI (one-time setup). You need Node.js 18+ first. If you don't have Node.js:
-
Log in with your ChatGPT account. Open a terminal and run:
codex login
A browser window will open. Sign in with the same ChatGPT Plus account you use at chatgpt.com. When done, credentials are saved to
~/.codex/auth.json. -
Configure the plugin. In Zotero → Preferences → llm-for-zotero:
- Set provider Auth Mode to
codex auth. - Set API URL to
https://chatgpt.com/backend-api/codex/responses. - Set Model to a Codex model (e.g.
gpt-5.4). - Click Test Connection to verify.
- Set provider Auth Mode to
Q: Is it free to use?
A: Yes, absolutely free. You only pay for API calls, if you choose to use a paid API provider. And now, you can even use Codex models (e.g. gpt-5.4) from your ChatGPT Plus subscription without an API key. If you find this tool helpful, please consider leaving a ⭐ on GitHub or buying me a coffee via Buy Me a Coffee / Alipay. Either way is greatly appreciated!
Q: Does this work with local models?
A: Actually, yes. As long as the local model provides an OpenAI compatible HTTP API that is compatible with the plugin, you can connect it by entering the appropriate API Base URL and secret key in the settings.
Q: Is my data used to train models?
A: No. Since you use your own API key, your data privacy is governed by the terms of the API provider you choose (e.g., OpenAI Enterprise terms usually exclude training on API data).
Q: If I have any questions, how to contact you?
A: Please feel free to open an issue on GitHub! I will try my best to help you.