@danielborek Do you suggest that main the Beaver output shows \cite{key} instead of [1] as citations? What about this instead: "Copy as Latex" option in addition to copy as markdown (which is the default). That would leave the UI unchanged (no \cite{key} in the visible output) but give you latex formatted output.
I was not suggesting that Beaver’s visible output should replace numeric citations with \cite{key}.
What I meant is indeed a copy/export preference: when a user copies Beaver output, citations could optionally be converted to a citekey-based format using the Zotero citekey already stored in metadata. That could support different targets, for example LaTeX (\cite{key}) or Pandoc/Markdown ([@key] / @key).
This is not an essential feature for me, and I would still verify citations manually, but it could be useful for workflows such as pasting Beaver output into Obsidian Markdown notes or LaTeX documents.
UI can stay unchanged, while the copied output uses a user-selected citation format.
EDIT: I shared here a long prompt on how to use extract info in relation to the research topic, but after a few days of using it, I see that it is better to use much shorter and focused questions, not necessarily spanning several subtopics.
@danielborek What I liked about your prompt is customizing AI workflows to your work. There are builtin defaults but it gets much more useful when you customize them.
@lboccia Not yet but almost done. Later this week or next week at the latest. With support for reading, editing and creating notes including citations, equations, user approval flows so you remain in control. Creating new notes already works now.
Joscha, thank you for Beaver — it has been a huge help for my workflow. As a private dentist with 35 years of manual Medline/PubMed searching behind me, I wanted to ask: are there any plans to integrate PubMed directly into Beaver? OpenAlex works well, but for medical topics the MeSH vocabulary and PubMed's filters are hard to replace. Native PubMed support would be a fantastic addition. P. S.: I just saw, that better integration between PubMed and Openalex is on its way: https://gemini.google.com/share/d08221172d3f
@milogies Glad to hear that Beaver is a huge help! Yes, I am interested in better supporting domain/discipline specific workflows starting with legal research, medical research and patent/intellectual property research. The goal would be to make Beaver more an expert in specific domains (in addition to general academic/research work). That might involve integrating with specific database (like Medline/PubMed), citation systems, zotero item types and other discipline-specific things that deviate from general-purpose academic/research practices.
If you (or others) are interested in sharing thoughts on that, feel free to send me an email. I am not an expert in any of these domains so it would be good to hear what is missing. For example, OpenAlex indexes PubMed so I think Beaver should already find PubMed articles. But maybe it's the focus, missing search fields etc
I recently created an open-access ebook using Beaver: https://www.logies.de/cliodynamics-buch.html Cliodynamics: An AI-Assisted Introduction Through Its Founding Journal, 2010–2026
Two points of feedback regarding the workflow:
Reference Handling: Beaver initially mixed up references. It required explicit prompting and pointing out the exact data field before Beaver integrated the Pandoc BibTeX references into the notes. Once exported to Markdown, Kimi 2.6/2.7 Code handled the build.
Editor Limitation: The lack of an "Undo" feature in the note editor makes correcting accidental edits stressful.
Despite these challenges, the project wouldn't have been possible without Beaver. Thank you for your work on it! Michael
If you are comfortable, it would be useful to have an example response that shows the issue with Pandoc BibTeX references. You can copy ids from the more options menu at the bottom and email it.
Joscha, it would be too time-consuming for me to try to reproduce the errors right now. But here was my workflow:
Beaver unfortunately cannot import entire journals (complete volumes, all issues), so I generated a list of DOIs for all 198 articles in the journal from Crossref using AI. I then fed them into Zotero in batches of 50, and Zotero was able to import almost all of them with full text because Cliodynamics: The Journal... is open access. I had to manually add a few full texts.
Then, in separate new chats, I handed 10 papers at a time over to Beaver and had it summarize each paper in an attached note of about 450 words (with clear instructions on note structure; the prompts were mostly AI-optimized as well).
Next, I asked Beaver/Kimi 2.6 to outline a 10–15 chapter, ~100-page book based on the 198 notes and the core thematic focus of the journal. It ended up being 12 chapters. During the chapter outline phase, Beaver was also prompted to specify which literature to use for each respective chapter.
Then I asked Beaver to write the book chapters. The writing took place in a single chat because Beaver suggested it would keep track of things better that way.
And then the book was done. ;-)
However, during a later review run, Beaver discovered that various citations were inaccurate, references were incorrect, and unused literature appeared in the bibliography of individual chapters. At that point, I also realized that Pandoc—which I had never worked with before—required citation keys from the corresponding Zotero field. So Beaver corrected all the errors (I hope) and inserted the Pandoc citation keys, including page numbers, into the notes.
After exporting all chapter notes, I handed them over to Kimi 2.6 (and later Kimi 2.7 Code) in my Linux VM, where Kimi set up the software environment for book publishing on its own. In the end, Kimi 2.7 Code apparently revised the formatting of all in-text citations, but that only cost time and money, not any effort on my part.
I also had the AI document the software environment setup; it's included in the technical appendix of my book. The book is free:
I initially had Kimi 2.6 translate it from English to German, but in too many places it didn't read very well (though still better than Kimi 2.7 Code). The translation by Claude Sonnet, on the other hand, turned out really well. About 10 USD.
The whole thing was a 3–4 day vacation project. That's all the time I had. Primarily, I wanted to write a book that I personally wanted to read, but which wasn't available on the book market. Besides, I learned to hate LaTeX 30 years ago. But with AI agents, I can now use both Linux and LaTeX productively. Michael
Joscha, thank you for Beaver，but It Just kinda hard for the people who don't use English very much,like me to use It。Can the extension be translated into Chinese
Upgrade Storage
What I meant is indeed a copy/export preference: when a user copies Beaver output, citations could optionally be converted to a citekey-based format using the Zotero citekey already stored in metadata. That could support different targets, for example LaTeX (\cite{key}) or Pandoc/Markdown ([@key] / @key).
This is not an essential feature for me, and I would still verify citations manually, but it could be useful for workflows such as pasting Beaver output into Obsidian Markdown notes or LaTeX documents.
UI can stay unchanged, while the copied output uses a user-selected citation format.
@lboccia Not yet but almost done. Later this week or next week at the latest. With support for reading, editing and creating notes including citations, equations, user approval flows so you remain in control. Creating new notes already works now.
É possível traduzir a extensão para português do Brasil?
https://www.logies.de/zotero-ki-plugins-2026.html
English:
https://www-logies-de.translate.goog/zotero-ki-plugins-2026.html?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=de#konfiguration
Highlights in my opinion:
- Getting Beaver to work with Kimi K2.6 through Openrouter.
- Custom Prompt for writing reviews automatically.
I am also trying to get this published in a German dental journal. I submitted the manuscript, co-written by Kimi Code and me, only yesterday.
P. S.: I just saw, that better integration between PubMed and Openalex is on its way: https://gemini.google.com/share/d08221172d3f
If you (or others) are interested in sharing thoughts on that, feel free to send me an email. I am not an expert in any of these domains so it would be good to hear what is missing. For example, OpenAlex indexes PubMed so I think Beaver should already find PubMed articles. But maybe it's the focus, missing search fields etc
I recently created an open-access ebook using Beaver:
https://www.logies.de/cliodynamics-buch.html
Cliodynamics: An AI-Assisted Introduction Through Its Founding Journal, 2010–2026
Two points of feedback regarding the workflow:
Reference Handling: Beaver initially mixed up references. It required explicit prompting and pointing out the exact data field before Beaver integrated the Pandoc BibTeX references into the notes. Once exported to Markdown, Kimi 2.6/2.7 Code handled the build.
Editor Limitation: The lack of an "Undo" feature in the note editor makes correcting accidental edits stressful.
Despite these challenges, the project wouldn't have been possible without Beaver. Thank you for your work on it! Michael
If you are comfortable, it would be useful to have an example response that shows the issue with Pandoc BibTeX references. You can copy ids from the more options menu at the bottom and email it.
Beaver unfortunately cannot import entire journals (complete volumes, all issues), so I generated a list of DOIs for all 198 articles in the journal from Crossref using AI. I then fed them into Zotero in batches of 50, and Zotero was able to import almost all of them with full text because Cliodynamics: The Journal... is open access. I had to manually add a few full texts.
Then, in separate new chats, I handed 10 papers at a time over to Beaver and had it summarize each paper in an attached note of about 450 words (with clear instructions on note structure; the prompts were mostly AI-optimized as well).
Next, I asked Beaver/Kimi 2.6 to outline a 10–15 chapter, ~100-page book based on the 198 notes and the core thematic focus of the journal. It ended up being 12 chapters. During the chapter outline phase, Beaver was also prompted to specify which literature to use for each respective chapter.
Then I asked Beaver to write the book chapters. The writing took place in a single chat because Beaver suggested it would keep track of things better that way.
And then the book was done. ;-)
However, during a later review run, Beaver discovered that various citations were inaccurate, references were incorrect, and unused literature appeared in the bibliography of individual chapters. At that point, I also realized that Pandoc—which I had never worked with before—required citation keys from the corresponding Zotero field. So Beaver corrected all the errors (I hope) and inserted the Pandoc citation keys, including page numbers, into the notes.
After exporting all chapter notes, I handed them over to Kimi 2.6 (and later Kimi 2.7 Code) in my Linux VM, where Kimi set up the software environment for book publishing on its own. In the end, Kimi 2.7 Code apparently revised the formatting of all in-text citations, but that only cost time and money, not any effort on my part.
I also had the AI document the software environment setup; it's included in the technical appendix of my book. The book is free:
https://www.logies.de/cliodynamics-buch.html
I initially had Kimi 2.6 translate it from English to German, but in too many places it didn't read very well (though still better than Kimi 2.7 Code). The translation by Claude Sonnet, on the other hand, turned out really well. About 10 USD.
The whole thing was a 3–4 day vacation project. That's all the time I had. Primarily, I wanted to write a book that I personally wanted to read, but which wasn't available on the book market. Besides, I learned to hate LaTeX 30 years ago. But with AI agents, I can now use both Linux and LaTeX productively. Michael