Lume

English | 中文

Warning

Lume is still a prototype. Expect bugs, rough edges, and breaking changes in both data flow and UX.

Lume is a local-first desktop app for reading and organizing academic PDFs. It combines a paper library, reader, annotations, notes, metadata enrichment, translation, AI summaries, and citation export in one Tauri application.

What It Does Today

• Local library with folders, subfolders, trash, rename, move, and search
• PDF import from file picker and drag-and-drop into the app window
• Multi-tab reading with lazy page loading, zoom presets, and in-document search
• Local annotations stored alongside PDFs and paper-level Markdown notes
• Metadata extraction from the PDF itself, then remote enrichment via arXiv, Crossref, and OpenAlex
• Metadata trace/report so you can inspect which provider filled which fields
• Citation generation and export
• Selection translation with google, bing web, or llm
• AI paper summary and annotation digest
• Native CLI for listing, searching, exporting, tagging, and opening items

Current Architecture

Frontend

• React 19 + TypeScript + Vite
• Tailwind CSS for styling
• src/App.tsx orchestrates the shell layout, drag-drop import, tabs, reader panels, and library interactions
• State is split into focused hooks:
  • useLibrary for library actions and Tauri command calls
  • useSettings for persisted settings and theme/font application
  • useFeedback for toast notifications
  • useI18n for runtime locale loading

Backend

• Tauri v2 + Rust
• src-tauri/src/lib.rs wires shared app state, plugins, CLI IPC, and command registration
• src-tauri/src/library_commands.rs owns library CRUD, search, notes, tags, export, translation, annotation sidecars, and settings persistence
• src-tauri/src/metadata_fetch.rs owns metadata parsing, provider orchestration, retries, caching, merge policy, and fetch reports
• src-tauri/src/pdf_handlers.rs owns PDFium-backed page rendering, text extraction, selection rects, and PDF-derived metadata hints
• src-tauri/src/cli.rs and src-tauri/src/cli_ipc.rs implement the native CLI and single-instance handoff to the GUI

PDF Pipeline

Lume currently uses a hybrid PDF stack:

• pdfium-render in Rust for raster rendering, text extraction, text-rect lookup, and metadata hints
• pdfjs-dist in the frontend for document/page caching and warmup

So the project is no longer accurately described as a pure “PDFium app” or a pure “PDF.js app”; both are part of the current runtime.

Storage Model

Application data lives under Tauri app_data_dir():

• library/ for imported PDFs
• trash/ for soft-deleted PDFs
• lume_library.db for SQLite-backed items, attachments, notes, tags, settings, and caches
• per-PDF annotation sidecars stored next to the imported file as .<filename>.Lume-annotations.json

The app is local-first: imported files are copied into Lume-managed storage instead of being referenced in place.

Metadata Retrieval Flow

Lume’s current metadata flow is modeled as a staged pipeline rather than a single lookup:

1. Parse candidate title/authors/year/DOI/arXiv ID directly from the PDF.
2. Run exact identifier lookups first (arXiv ID, Crossref DOI, OpenAlex DOI).
3. If metadata is still incomplete, run fuzzy title searches (OpenAlex, Crossref, arXiv) with title variants plus author/year scoring.
4. Merge results field-by-field with provider priority and preprint-aware rules.
5. Cache the result and persist a metadata fetch report for later inspection in the UI.

This is designed to improve noisy PDF imports and avoid letting preprint metadata overwrite a confirmed venue/publication result.

AI and Translation

• AI summaries and LLM translation use a user-configured OpenAI-compatible completion endpoint
• Non-LLM selection translation can use:
  • google via the public web endpoint
  • bing via the Bing Translator web flow
• llm translation requires the AI endpoint settings to be configured

Quick Start

Local Development

npm install
npm run tauri dev

CLI

Examples:

Lume list
Lume list --json
Lume search "transformer" --json
Lume export --format bibtex -o refs.bib
Lume open /absolute/path/to/paper.pdf

During development:

npm run cli:list

Build

npm run tauri build

Detailed packaging and release notes are in docs/build-and-release-guide.md.

Tech Stack

• Desktop shell: Tauri v2
• Frontend: React 19, TypeScript, Vite
• Styling: Tailwind CSS
• Backend: Rust
• Database: SQLite via rusqlite
• PDF engines: pdfium-render and pdfjs-dist
• Network metadata sources: arXiv, Crossref, OpenAlex

Project Status

The app is moving fast. The current codebase already contains the foundations for a serious local research workflow, but it is not stable yet. If you are evaluating the project, the best framing is:

• usable for development and experimentation
• not yet safe to trust as your only literature manager

For roadmap context, see docs/zotero-gap-analysis.md.