Writing a paper is mostly not writing. It is fighting BibTeX, hunting for the one citation you half remember, rebuilding a figure for the fourth time, reformatting for a conference template that uses the wrong class file, and then doing the whole thing again for the poster and the slides.
He got tired of it, so he built Bibby AI, a LaTeX-native AI auto-scientist that lives in one editor and does the parts that are not actually thinking.
Right now it is used by 10,000+ researchers across 50+ universities in 50+ countries. Here is what it actually does.
What "auto-scientist" means (and what it does not)
Auto-scientist is a loaded phrase, so let me be direct about it. Bibby is not a magic button that produces a Nature paper from a title. The full auto-scientist loop breaks in real science on data scarcity, missing experimental common sense, and the gap between "statistically significant" and "actually relevant."
So Bibby owns the parts where that judgement is real and repeatable: write, cite, figure, review, present. You keep the science. It removes the friction around it.
LaTeX writing without babysitting the compiler
Bibby is a full agentic LaTeX editor. You can:
Write in raw LaTeX with an AI toolbar sitting right next to the document
Or use Rich Text mode, write like it is a Word doc, and let Bibby handle the LaTeX underneath
Ask the agent to rewrite a paragraph, tighten an abstract, or fix a section, in place
If you have ever used Overleaf, the mental model is the same, except the assistant understands the document instead of just autocompleting brackets. It is built as a real Overleaf replacement, not a plugin bolted on top.
latex
% Ask Bibby: "make this intro tighter and add a citation for the claim"
% It rewrites the text AND drops in the real \cite{} + the .bib entry
Citations that are real, searched live
The worst failure mode of AI writing tools is confident fake references. Bibby does live citation search inside the editor. You describe the claim, it finds the actual paper, and it inserts the \cite{} plus the BibTeX entry for you.
No copy-pasting DOIs. No hallucinated authors. No "et al., 2019" that does not exist.
Figures and diagrams, generated in the paper
This is the part people do not expect from a writing tool. Bibby ships Academic Diagrams, agentic figure creation with:
6,000+ scientific icons (Bioicons + SciDraw)
AI-generated diagrams with Restyle, Edit, Methodology, and Architecture modes
One-click Insert into paper, which injects the figure straight into your LaTeX
Think BioRender energy, except the output lands in your document as a figure instead of a PNG you have to wrangle. You can generate a methodology diagram or an architecture figure from a prompt and drop it in without leaving the editor.
Posters and presentations, from the paper you already wrote
Once the paper exists, the poster and the slides are usually a second full project. Bibby collapses that. The agent turns your existing document into conference posters and presentation decks, so you are editing an output instead of starting from a blank template at 2am the night before the deadline.
The whole point: your figures, citations, and text are already in one place, so generating the poster is a transform, not a rebuild.
The pile of LaTeX tools around it
The editor is the core, but a lot of the daily value is in the small tools:
LaTeX to Word converter at trybibby.com/latex-to-word. Real client-side parser with proper OMML math and a clean docx export, for the co-author or reviewer who refuses to touch LaTeX.
The Paper Clock at trybibby.com/conference-deadlines. Live countdowns for ~80 venues across CS, neuro, physics, and math. There is a Chinese version too.
AI paper reviewer, so you get a critical read before an actual reviewer does.
Research Companion and LitLens Chrome extensions for lab search and literature, powered by ROR and OpenAlex.
A searchable research-software directory for finding the right open tool (DeepLabCut, CaImAn, SpikeInterface, NWB, and more).
Who this is for
If you write papers, grants, theses, or posters and you touch LaTeX at any point, this is built for you. It is especially aimed at researchers who want the AI to do the mechanical 60% without pretending it can do the 40% that is your actual contribution.
It also plugs into university SSO. The parent entity is now a UK Federation member via Jisc, which unlocks SAML and Shibboleth login across a large chunk of the global research federation network. So if your institution supports it, you may be able to log in with your university account.
Try it
It is live at trybibby.com. There is a free tier, so you can open the editor, generate a figure, and see the citation search work before deciding anything.
If you are a researcher or a dev who has suffered through LaTeX tooling: what is the one part of paper production you would automate first, the citations, the figures, or the reformatting? Tell me in the comments, that is genuinely what drives the roadmap.
