简体中文 | English
RubyCite is a desktop app for macOS Linux and Windows that turns your computer into a Research IDE: import papers, read and search them locally, ask questions across your library, draft with citations, and finish in Word — without handing your PDF collection to a cloud service by default.
Everything lives on your machine: converted texts, library records, vector index, notes, and project manuscripts.
RubyCite fits researchers and graduate students who:
- collect English journal PDFs (and bibliographic files) for literature reviews and thesis writing;
- want semantic search and Q&A over their own papers, not the open web;
- need sentence-level citation suggestions while drafting;
- write long manuscripts in Markdown and export to Word with EndNote-style or CSL-formatted references;
- prefer local LLMs (built-in or Ollama) or their own API keys.
Build a searchable collection, then use it while you write elsewhere.
Import PDFs or RIS / BibTeX / EndNote .enw / Markdown
→ Organize in groups; edit metadata; preview converted text
→ Build index on selected papers
→ Ask questions in the AI panel (optional)
→ Paste a draft into Citation assist → get {Author, Year} markers
→ Export RIS → Word / Zotero / EndNote
Import papers
→ AI Summary: five structured cards per paper + your reading notes
→ Review assist: feed multiple papers → generate a literature review
→ Export RIS or copy the review for Word
Open a project folder and treat RubyCite like a lightweight writing IDE:
New project from template (paper / review / grant / final report / empty)
→ Edit .md manuscripts with CodeMirror (syntax highlight, @ citation insert)
→ Check Outline, Preview, Bibliography, Dashboard, Problems
→ Export Word (EndNote markers, static CSL text, or RubyCite Word controls)
→ Optional: Word add-in to refresh citations in Word
All three workflows share the same library and the same project bibliography.
- Converts born-digital PDFs with selectable text (publisher articles from Elsevier, Springer, IEEE, etc.).
- Extracts body text, tables, figures, and layout-aware reading order (including two-column articles).
- Rejects scanned or image-only PDFs when text cannot be extracted reliably.
- Also imports RIS, BibTeX, EndNote
.enw, and existing Markdown files. - You can create references manually without a PDF.
- A central paper table with full-text search (SQLite + FTS).
- Manual groups and smart groups (e.g. has PDF, has AI summary, year ≥ 2020).
- Rich metadata editing: title, authors, year, journal, DOI, cite key; refresh from Crossref; CSTR verification.
- Find and merge duplicate entries.
- Paper preview of converted Markdown (headings, tables, figures, reference list line by line).
- Separate tabs for AI summary cards, your reading notes, figure list, and metadata / CSL preview.
- Right-click a row for quick access to metadata, preview, index actions, or delete.
- Copy formatted bibliographies (GB/T 7714, APA 7, IEEE, author–year) and import custom CSL styles.
- Builds a local vector index with the built-in multilingual-e5-small embedding model.
- You choose which papers to index; chunk size is configurable in Settings (rebuild after changes).
- Ask natural-language questions over indexed papers; answers follow the language of your question.
- See index status, indexed file list, and remove individual papers from the index.
- Paste a full draft; RubyCite splits it into sentences and searches the index for each.
- Up to three source passages per sentence; same paper is merged into one
{Author, Year}marker. - View which sources matched each sentence; copy the annotated draft for Word.
- Requires a built index first.
- Generates five AI cards: Scientific question, Innovation, Key findings, Limitations & outlook, Future research directions.
- Saves a lightweight
.digest.mdlinked to the full paper; you can add personal reading notes on a dedicated tab. - Export digest (e.g. for Obsidian).
- Output language is selectable.
- Check papers → Feed to model → choose template: journal article, thesis, or survey report.
- Write a custom prompt; generate a long-form review in the AI panel.
- Save or copy the review; export RIS for reference managers.
- Explorer for your project folder; create files and folders; open multiple manuscript tabs.
- CodeMirror 6 editor with Markdown highlighting.
- Project bibliography in the sidebar — link library papers to the current project.
- Two citation syntaxes in one document:
[@cite_key](structured) and{Author, Year}(Word/EndNote). - Type
@to search the library and insert cite keys.
Workbench tools for the open manuscript:
| Tool | Purpose |
|---|---|
| Outline | Jump through headings |
| Preview | Render manuscript with formatted citations, KaTeX math, Mermaid diagrams |
| Bibliography | List unresolved or project references |
| Overview / Dashboard | Word counts, bibliography size, index status |
| Problems | Unresolved citations, orphan bib entries, missing images — click to jump |
| Citation network | Which library papers appear in which manuscripts |
| Gap analysis | LLM report on themes and gaps in the project bibliography |
| AI pre-review | Scores (innovation, evidence, coverage, writing) plus weakness list |
Editor assist (needs LLM in Settings; evidence and suggestions need index):
- Hover
[@cite_key]for details; F12 jumps to the library entry. - Suggested citations for the current paragraph (after index is built).
- Cmd+K custom rewrite; Cmd+L ask about selection.
- Right-click: Rewrite, Academic tone, Expand, Shorten, Add citation, Find evidence.
- Find evidence groups indexed papers as supporting, contradicting, or neutral for a selected claim.
- Tab at paragraph end for ghost-text continuation (accept with Tab).
Export to Word (.docx):
- EndNote mode —
{Author, Year}markers for “Update citations” in Word. - Static CSL — in-text and reference list text in your chosen CSL style.
- RubyCite mode — Word content controls refreshable via the add-in.
- Optional
reference.docxtemplate for fonts and heading styles (bundled Pandoc).
Export can be blocked when the project scan reports unresolved citation errors (you may override with confirmation).
Start faster with scaffolds:
- Empty — metadata only
- Paper — manuscript + figures folder
- Review — survey + themes notes
- Grant — NSFC-style proposal sections (background, gap, question, methods, innovation, plan, budget) + experiment folder
- Final report — closure report structure
Create a new folder from template or apply a template to an open workspace.
- Local HTTPS connection to RubyCite on your computer (desktop app must be running).
- Task pane: search library, insert citations, format bibliography.
- One-time certificate trust and manifest install from the Tools menu.
- Support Libreoffice and Onlyoffice on Linux
- LLM providers: built-in llama.cpp (local
.gguf), Ollama, LM Studio, vLLM, OpenAI-compatible APIs (OpenAI, DeepSeek, Groq, OpenRouter, Moonshot, Silicon Flow, Azure, and more). - Six interface languages: English, 中文, 日本語, Français, Español, Deutsch.
- Color themes: default, blue light, warm light, blue dark.
- Crossref email for DOI metadata enrichment (optional, needs network).
- Converted Markdown,
library.db, vector index, RIS export, reviews, and workspace files stay under your RubyCite data folder or workspace folder. - AI runs locally or through APIs you configure — RubyCite does not require a vendor cloud account.
- The Word add-in talks to localhost only; it does not upload your library to RubyCite servers.
| Requirement | Detail |
|---|---|
| PDFs | Text must be selectable (not scan-only) |
| Index / Citation assist | Build index on selected papers first |
| AI features | Configure an LLM in Settings → Test connection |
| Word add-in | Trust certificate; install manifest; keep RubyCite running |
| Network | Only for Crossref, cloud LLMs, or optional model downloads |
For step-by-step use, open Help → User guide inside the app.
| Resource | |
|---|---|
| In-app User guide | Full manual |
| About | Version and legal notices |
| EULA.md | License terms |
| COPYRIGHT.md | Copyright |
Contact: chemiker2010@gmail.com
RubyCite 2.0 — Copyright © 2026 Hongtao Lu. All rights reserved.