Star0
0
Code
Code
Issues
Pull Requests
Discussion
Wiki
Analysis
Star0
0
LNN0212LNN0212Initial Project Submission
f73b6250Add on 2 days agoHistory

Usage Guide

简体中文

This guide walks through the main workflow: discover papers, organise them into projects, analyse them, and read them with evidence-aware AI support.

1. Start and sign in

Start with start.sh / start.bat, then open:

http://127.0.0.1:8765

Enter the access code shown in the launcher console. The default bind address is 0.0.0.0 so other devices on the LAN can connect with the same code; change it to 127.0.0.1 in Settings → Network if you want local-only access.

2. Navigation

Tab Purpose
研究工作台 / Workspace Home dashboard: projects, recent reading, to-read, recent analyses, notes and progress
论文检索 / Search arXiv search and download
我的论文库 / Library Downloaded arXiv PDFs
期刊检索 / Journal search Journal screening and acquisition workflow
科研课题 / Projects Project workspace (main working area)
本地论文 / Local papers Unified local library across arXiv and journal papers
设置 / Settings AI, network, downloads, interface, journal workflow

The dashboard is the default view, but search remains available at any time.

3. Search arXiv

  1. Open 论文检索.
  2. Enter a topic and optional keywords.
  3. Keep Use AI enabled to let the configured model build the query; disable it for plain keyword search.
  4. Choose result count and sort order.
  5. Search, select papers, then Download.

Downloads run as durable jobs. You can close the page, reopen it, or restart the app; unfinished jobs are restored on startup. Progress is shown in the task panel.

AI query generation falls back to a plain query when the model is unavailable or returns invalid syntax.

4. Download and local library

  • PDFs are stored under data/pdfs/ (arXiv) and data/review/pdfs/ (journals).
  • The download engine retries mirrors, resumes partial files, checks the PDF signature and size limit.
  • Optional Playwright fallback can be enabled in Settings for restrictive networks.
  • 我的论文库 lists downloaded arXiv files with search and sorting.
  • 本地论文 merges arXiv and journal papers, shows analysis state and provides reader/analysis actions.

If a paper has no PDF, use Download or import a PDF from the journal workflow.

5. Journal workflow

Open 期刊检索:

  1. Choose a journal (built-in or custom ISSN).
  2. Set the result count and optional online-date window.
  3. Edit research goal / relevance criteria.
  4. Search, then screen the results (rule-based score plus optional AI review).
  5. Select papers and add them to the processing queue.

Queue states include: pending screen, pending download, manual action required, downloaded, analysing, completed, failed and skipped.

For publisher pages with a login/captcha/paywall, Paper Atlas opens a real Firefox window and waits for you to save the PDF. It never bypasses access controls.

Use Settings → Journal workflow to configure the Elsevier key, contact email, manual download folders, analysis prompt and auto-resume behaviour.

6. Create a research project

Open 科研课题:

  1. Click 新建课题 / New project.
  2. Fill in name, description, background, objectives, keywords, research directions, icon, colour and research status.
  3. Save.

A default project is created automatically when migrating an older single-goal installation. Rename it or split it into multiple projects at any time.

Project list

  • Search by name, description or keywords.
  • Show/hide archived projects.
  • Each row shows paper/core/to-read counts.
  • A project can be archived and restored; deletion requires confirmation and shows how many links, questions, notes and progress entries will be removed.

Deleting a project does not delete the papers themselves.

7. Project workspace

Open a project to see its workspace with eight tabs.

Overview

Edit name, description, background, objectives, directions, keywords, icon, colour and research status. Shows paper/question/note counts.

Research Questions

  • Add a question with description, hypothesis, keywords, status and linked papers.
  • Edit or delete existing questions.
  • Use AI 辅助拆解 / AI decompose to generate candidate questions.
  • AI candidates are not saved automatically; each one must be adopted by the user.

Literature

  • Add papers by DOI from the existing library.
  • Remove a link without deleting the paper.
  • Mark core / 重点 and switch unread / reading / read.
  • Add personal evaluation, a 0–5 rating and tags.
  • Filter by role, reading status, year, tag and analysed state.
  • 阅读 / Read opens the PDF reader.

A paper can be linked to multiple projects. The papers row and PDF file are shared, not duplicated.

Analysis

  • Lists every project paper with its analysis badge.
  • View analysis opens the existing analysis modal.
  • Analyse queues a full-text analysis job; a local PDF is required.

Notes

  • Create, edit and delete Markdown notes attached to the project (and optionally a paper/page).
  • Notes created by the AI assistant can be edited like any other note.

Progress

  • Record logs and milestones with an optional 0–100% progress value.
  • Reports (kind = report) are shown under Review.

Review

  • View and create report-style Markdown documents for the project.
  • Use the AI outline action and save the result as a report; then refine it manually.

AI assistant

Actions:

  • Summarise current progress
  • Summarise research directions
  • Recommend reading
  • Refine research questions
  • Suggest research gaps
  • Generate a review outline
  • Ask a custom question

The assistant reads the project description, goals, questions, linked papers, existing analyses and notes. It never writes to the database automatically. Citations are restricted to DOIs present in the project; invalid citations are removed and reported. An answer can be saved as a note or report, which you can edit.

8. PDF reader

Open a paper from:

  • Project → Literature/Analysis → Read
  • Local papers → Read
  • Library → Read

Layout

Left: PDF. Right: AI analysis, notes and annotations.

Reading controls

Control Action
Page input / arrows Jump between pages
− / + / percentage Zoom
适宽 / 整页 Fit width / fit page
单页 / 连续 Single-page or continuous lazy rendering
目录 PDF outline/heuristic sections
搜索 Search cached page text
书签 Toggle a bookmark on the current page
课题 dropdown Choose a project context
标为已读 / 待读 Update project reading status

Reading position, zoom, mode and side tab are restored per paper after closing and reopening.

AI analysis panel

The panel shows up to twelve modules and marks missing content honestly. Each module can be expanded/collapsed. Results that include evidence show a clickable evidence card with:

  • PDF physical page
  • section and figure (if available)
  • quote
  • verification status

Clicking evidence jumps to the page. When the browser text layer can match the quote, the source is highlighted; otherwise the reader goes to the page and shows the quote without inventing a highlight.

Select text

Select text in the PDF, then choose:

  • Explain
  • Summarise
  • Translate
  • Explain formula
  • Explain method
  • Relation to research goal
  • Highlight / underline / comment / key evidence
  • Save as project note

AI answers are shown in the side panel and can be saved as a note. The selection action never modifies the paper or existing notes automatically.

Annotations and notes

  • Highlights/underlines/comments are stored with normalised coordinates so they survive zoom and reopen.
  • Bookmarks are stored per page.
  • Key evidence is highlighted in red and listed in the annotation panel.
  • Notes attached to a paper can be created, edited and deleted.
  • Saving a selection as a project note stores it under the currently selected project.

Limitations

  • Printed page numbers are not guessed; the UI marks them as unknown.
  • Scanned PDFs need OCR and are not supported.
  • Cross-page selections are saved against the starting page.
  • If analysis was produced by an older schema, some panels may be empty; the reader does not infer missing content.

9. AI configuration

Open 设置 → AI:

DeepSeek

  • Provider: DeepSeek
  • Base URL: https://api.deepseek.com
  • Model: deepseek-chat (or another available model)
  • API key: your key

Ollama (local)

  1. Install and start Ollama.
  2. Pull a model, e.g. ollama pull qwen2.5:7b.
  3. Select Ollama, native URL http://127.0.0.1:11434, model name.
  4. No API key is required.
  5. Use Test to verify the connection.

OpenAI-compatible endpoint

Choose OpenAI-compatible, enter the base URL, model and key. This works with LM Studio, vLLM, OneAPI and hosted gateways.

The key is stored only in data/config.json and masked in the UI. Environment variables take precedence.

10. Settings

Section What you can change
AI Provider, model, base URL, key, enabled state, test connection, Ollama model pull
Network Host, port, access code, language, theme
Downloads Max PDF size, concurrent jobs, resume, browser fallback
Interface Language and theme
Journal workflow Journal, research goal, relevance threshold, AI screening, analysis prompt, Elsevier, Firefox, manual folders
Diagnostics Version, Python, data folder, PDF folder, optional modules, health, download-host test
About System information and service status

Host/port changes require a restart. Other AI/download settings apply to new requests.

11. Backup and data management

python scripts/backup_data.py       # DB + config + library + analysis copies
python scripts/verify_data.py       # integrity, counts, PDF presence
python scripts/migrate_schema.py    # apply pending schema migrations

Backups are written to data/backups/ and exclude the Firefox profile. See INSTALL.md for restore and migration details.

12. Keyboard and browser notes

  • Ctrl/Cmd + Enter submits the search form.
  • Esc closes the task panel or toast; in the reader it closes the reader (or selection menu).
  • Arrow keys navigate pages in the reader.
  • The UI is fully local: no CDN fonts, icons or scripts are required.
  • Use a current Firefox/Chrome/Edge for the PDF reader and text layer.

13. Common workflows

Paper → project → analysis → reading

  1. Search/download or find a local paper.
  2. Create a project and link the paper.
  3. Run analysis from the project Analysis tab.
  4. Open the reader and click evidence to check the source.
  5. Highlight key sentences and save notes.
  6. Use the AI assistant to summarise progress and propose gaps.

Journal screening

  1. Configure the journal and research goal.
  2. Search a date window.
  3. Review relevance scores; re-screen with AI if needed.
  4. Queue selected papers, complete manual publisher steps if prompted.
  5. Analyse and read the completed items.