Skip to content
SakethGGPublic

About

Frontline stands at the front line of research: it takes the daily barrage of new papers, so you only get the ones that matter, first.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

Β 

History

29 Commits

Folders and files

Repository files navigation

Frontline: the research radar that reads the papers for you

macOS and Windows Python 3.10+ Runs locally MIT license

Stop drowning in arXiv. Frontline finds the new papers in your field, reads them,
and tells you which ones are worth your time, all from one desktop app.


Why Frontline?

Keeping up with research is a part-time job. Hundreds of papers land every week, alerts are noisy, and the one paper that matters is buried on page three. Frontline does the digging so you can do the thinking.

Without Frontline: many tabs, noisy alerts, lost PDFs. With Frontline: one click, only new papers, AI summaries and scores.

  • ⚑ One click, three databases. Press Sweep now and Frontline searches arXiv, Semantic Scholar and OpenAlex at the same time.
  • 🧹 No repeats, no noise. The same paper from different sources is merged into one, papers you've already seen are skipped, and off-topic results are dropped.
  • 🧠 Read 20 papers in 2 minutes. AI writes a short briefing card for each paper (TL;DR, key points, method, results, limitations) and scores how relevant it is to you, from 0 to 10.
  • πŸ”¬ Follow the labs you care about. Add a professor by name, and you'll see every paper their lab puts out.
  • πŸ“š Everything else in one place. A paper library, a deadline tracker (yours and conferences'), an ideas notebook and a project board, all connected.
  • πŸ”’ Your data stays yours. Everything is stored on your computer. Your PDFs are plain files in a folder you choose.

How it compares

Frontline Google Scholar alerts arXiv email digests Reference managers
Finds new papers in your field βœ… βœ… βœ… ❌
Searches several databases at once βœ… βž– ❌ ❌
AI summary for every paper βœ… ❌ ❌ ❌
Ranks papers by relevance to you βœ… ❌ ❌ ❌
Follow a lab, not just an author profile βœ… βž– ❌ ❌
Chat with a paper's full PDF βœ… ❌ ❌ βž–
Personal + conference deadlines, calendar export βœ… ❌ ❌ ❌
Ideas β†’ proposals β†’ projects βœ… ❌ ❌ ❌

A morning with Frontline

  1. β˜• Open the app. Your morning sweep has already run (if you scheduled it), and a notification tells you what it found.
  2. πŸ‘€ Skim the cards. The best papers are at the top with a green score. Drag the relevance slider to hide the rest.
  3. πŸ“– Go deep on the one that matters. Click Deep dive and AI reads the full PDF, then gives you the method, the real numbers and a read/skip verdict. Got a question? Ask the paper.
  4. πŸ’Ύ Keep it. Save to library puts the PDF into your library folder, already tagged and summarized.
  5. πŸ’‘ Act on it. Turn the spark into an idea, the idea into a proposal, and the proposal into a project with tasks and a target conference.

Everything in one window

Six sections: Home, Radar, Library, Deadlines, Ideas, Projects

Section What it does for you
🏠 Home Today at a glance: a focus timer, next deadlines, tasks due this week, the best new papers, your reading queue and recent ideas.
πŸ“‘ Radar Your feed of fresh papers, with AI summaries, scores and a sweep digest.
πŸ“š Library Your PDFs in a folder you choose (any local folder, or a synced one), tagged by your interests automatically.
πŸ—“οΈ Deadlines Live countdowns to your own deadlines and your venues', from a catalog of ML, vision, NLP and robotics conferences and workshops.
πŸ’‘ Ideas Capture ideas, grow them into proposals and roadmaps, and pressure-test them with AI.
βœ… Projects A mission board (tasks earn XP, levels, streaks and badges) and backlog for each project, counting down to its target venue.

How it works

Search, merge, filter, read, notify

No AI key? Frontline still works. Cards fall back to extracts from the abstract and keyword-based scores.


Get started in 2 minutes

You need Python 3.10+ and, for the AI features, an Anthropic (Claude) or OpenAI API key.

🍎 macOS

git clone https://github.com/SakethGG/Frontline.git ~/Projects/Frontline
cd ~/Projects/Frontline
./install-desktop.sh        # one time: sets up .venv and creates ~/Applications/Frontline.app

Put your ANTHROPIC_API_KEY (or OPENAI_API_KEY) in .env, then open Frontline from Spotlight (Cmd+Space, type "Frontline"). To keep it in the Dock, right-click its icon while it's running and choose Options β†’ Keep in Dock.

πŸͺŸ Windows

git clone https://github.com/SakethGG/Frontline.git
cd Frontline
.\install-desktop.ps1       # one time: sets up .venv and adds Desktop + Start-menu icons

Put your API key in .env, then double-click Frontline on your Desktop.

🌐 Linux, or any browser

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
cp .env.example .env        # then add your API key
.venv/bin/python -m uvicorn frontline.app:app --port 8000

Open http://localhost:8000. Press Ctrl+C to stop.

Your first sweep

  1. Go to Radar and add a topic (for example, "efficient LLM inference").
  2. Optionally, click + Researcher to follow a lab.
  3. Press Sweep now. A minute later you'll have your first stack of briefing cards.

Features in detail

πŸ“‘ Radar: papers, summaries and researchers
  • One-click sweep across three free scholarly APIs. Results are merged by arXiv ID, DOI or title, and papers you've already seen are skipped, so each sweep shows only new work.
  • Topics with subtopics. It starts empty: add any field from the app. Each topic has search phrases, optional arXiv categories and a look-back window.
  • AI briefing cards: TL;DR, 3–5 key points, problem / method / results / why it matters / limitations, paper type, and a relevance score with a one-line reason. Use the min-relevance slider to hide noise.
  • Sweep digest: headline must-reads, themes and "worth watching" across everything found.
  • Deep dive: AI reads the full PDF and writes a dense breakdown covering method, setup, the actual numbers, limitations, ideas worth stealing and a read/skip verdict.
  • Ask the paper: a chat about any paper, grounded in its PDF. The PDF is prompt-cached, so follow-up questions cost less.
  • Researchers you follow. Click + Researcher, type a professor's name, and pick the right person from the candidates (institution, h-index and research areas tell namesakes apart). Their feed shows every paper they co-author, which in practice is their lab's output. It draws on OpenAlex (by author ID, plus same-name papers at their institution, since author databases often split one person into several profiles) and on arXiv (by name, within the categories they publish in) for the freshest preprints. Following someone backfills the last 6 months right away. AI scores each paper against the lab's research areas, so a paper wrongly attributed to a namesake scores low and is hidden.
  • Reading workflow: star papers, track unread, filter by topic, subtopic or sweep, full-text search, one-click BibTeX, code-link detection, and Markdown export of any view.
  • Notifications: desktop notification when a sweep finishes, optional phone push via ntfy.sh (no account needed), and an optional email digest.
  • Scheduled sweeps (FRONTLINE_AUTO_SWEEP_HOURS), so new papers arrive without you pressing anything.

Navigating: topic pills β†’ New / All / Starred tabs β†’ optional subtopic chip β†’ click any card to open the reader panel. Its β€Ή β€Ί buttons step through papers without closing it. Opening a paper marks it read, so New works like an inbox you can clear.

πŸ“š Library

You pick any folder on your computer. Put it inside Google Drive for desktop, Dropbox or OneDrive if you also want it synced across devices. Frontline keeps plain PDF files there:

  • Sub-folders are collections, and you browse them like folders: clicking a paper opens the PDF, Details opens its notes.
  • Files you drop into the folder from anywhere show up on the next scan. Scans run at startup and when you open the Library.
  • Moving or renaming a file keeps its notes, because files are matched by content.

Every PDF is identified from its arXiv ID or DOI when it has one, so you get the exact title, authors and abstract. It's then tagged against your interests and scored for fit. By default your interests are your Radar topics and subtopics, and you can edit them. With an AI key, tagging is one short call per new paper; without one, it uses keyword matching.

  • Filtering: search, filter by interest, or show starred papers. Sort by relevance to you, date added, year or title.
  • Paper drawer: take notes, edit tags, move a paper between collections, copy its BibTeX, and run Deep dive or Ask on your local PDF.
  • Adding papers: drag and drop, or + Add PDFs. On any Radar paper, – Save to library downloads its PDF into the folder and carries over its AI card.
  • Removing papers: a removed PDF moves to a hidden .frontline-trash folder inside the library, so nothing is lost.

Google Drive tip (if your library is in Drive): with Drive for desktop set to stream files, Frontline reads each new PDF once to index it, and Drive then downloads it. To keep a large library available offline, mark the folder Available offline.

πŸ—“οΈ Deadlines: personal, conferences & workshops
  • One timeline: your own deadlines and your tracked venues' deadlines count down together. Filter to Personal or Conferences.
  • Personal deadlines: + Deadline (or n) adds anything with a date: a thesis draft, a grant report, a visa renewal. The time is optional (no time means the end of that day), and you can link it to a project. Tick it off when it's done; open ones stay on the timeline until you do.
  • Catalog: browse the community-maintained huggingface/ai-deadlines catalog. It covers ML, vision, NLP, robotics and more, with every deadline converted from AoE/PST/UTC into your local time.
  • Focus areas: pick yours to get open deadlines suggested.
  • Tracking: tracked venues stay up to date. When the catalog announces a series' next edition, it's added for you.
  • Manual entries: add workshops, journals or anything missing by hand. If you edit a catalog venue's dates, later updates leave it alone.
  • Calendar export: πŸ“… Export calendar downloads an .ics file with every open deadline, personal and conference (reminders a week and a day before), the conference dates and your dated tasks. Import it into Google Calendar, Outlook or Apple Calendar.
πŸ’‘ Ideas
  • Capture: type an idea and press Enter. Proposals and roadmaps start from a template.
  • Develop: write Markdown notes (they save as you type), jot small quick notes beside them (colour-coded sticky notes for passing thoughts, edited in place), add tags and a milestone checklist, link the papers an idea builds on, and move ideas through Spark β†’ Exploring β†’ Developing β†’ Active β†’ Parked on a drag-and-drop board.
  • AI assistant (API key needed):
    • Pressure-test gives strengths, weak spots, a novelty check against the linked papers, and next steps.
    • Brainstorm suggests six directions.
    • Draft proposal writes a full proposal outline.
    • Plan milestones adds a realistic roadmap.
  • πŸš€ Start a project: turns an idea into a project. Its open milestones become tasks: the first goes on the To-do board, the rest go to Future.
βœ… Projects
  • Mission board: Missions / In action / Completed, with drag and drop, due dates (overdue ones are flagged) and priorities.
  • Missions & XP: every task is a mission worth XP (Easy 10, Standard 20, Boss 40, plus 10 for finishing by its due date). XP levels you up through ranks from Recruit to Legend. A daily goal (adjust it with βˆ’ / +) pays a 30 XP bonus, a streak counts the days in a row you finish something, and badges unlock as you go. Finishing a mission pops confetti. Everything is worked out from your tasks, so un-ticking or deleting a finished task takes its XP back.
  • Future tab: for things you'll do later, so the board stays focused.
  • Target venue: a project can point at a tracked conference, and its header counts down to the next deadline.
  • Inbox: collects tasks with no project. All tasks shows everything.
⌨️ Keyboard shortcuts
Keys Action
h r l c i p Jump to Home / Radar / Library / Deadlines / Ideas / Projects
n New idea or task
f Start / pause the focus timer (on Home)
/ Search the current section
? Show all shortcuts
j / k Next / previous paper (Radar)
Enter / Esc Open / close a paper
s / u Star / toggle read
d Deep dive
1 2 3 Switch tabs
Shift+S Sweep now

Moving to a new computer

There's no account or cloud, so your data travels as one file. Open βš™ Settings β†’ Your data:

  1. ⬇ Back up to a file saves frontline-backup-<date>.frontline. It holds your topics, followed researchers, papers, summaries, stars and read marks, library index, ideas, projects, tasks, deadlines, conferences and UI preferences. Tick Include library PDFs if your PDFs aren't already in a synced folder (Google Drive, Dropbox, …).
  2. Install Frontline on the new computer, open βš™ Settings β†’ ⬆ Restore from a backup… and pick the file. Everything comes back as it was. Restored PDFs go into your library folder, or ~/Frontline Library if there isn't one yet.

API keys aren't in the backup: copy your .env file across yourself. The .frontline file is a plain zip (manifest.json, frontline.db, optional library/). Restore also accepts a bare frontline.db, so copying data/frontline.db by hand works too.

Configuration

All settings live in .env (copy it from .env.example). Only an API key is needed to get going.

All settings
Variable What it does
ANTHROPIC_API_KEY Enables Claude summaries, deep dives, Q&A and digests
FRONTLINE_MODEL Claude model (default claude-opus-5-5)
OPENAI_API_KEY Use OpenAI instead of Claude for the same features
OPENAI_MODEL OpenAI model (default gpt-5-mini)
FRONTLINE_LLM claude, openai or auto (default). auto uses OpenAI only when OPENAI_API_KEY is set and ANTHROPIC_API_KEY isn't
FRONTLINE_SUMMARY_EFFORT Effort for card summaries (default low, which keeps sweeps fast and cheap)
FRONTLINE_DEEP_DIVE_EFFORT Effort for deep dives (default medium)
FRONTLINE_MAX_PAPERS_PER_TOPIC Cap on new papers summarized per topic per sweep (default 40)
FRONTLINE_CONCURRENCY How many papers are summarized at once (default 4)
FRONTLINE_RESEARCHER_LOOKBACK_DAYS How far back a followed researcher's feed goes (default 180)
FRONTLINE_MAX_PAPERS_PER_RESEARCHER Cap on new papers summarized per researcher per sweep (default 60)
FRONTLINE_AUTO_SWEEP_HOURS e.g. 12 runs a sweep twice a day while the app is open
NTFY_TOPIC Push notifications. Install the ntfy app, subscribe to a hard-to-guess topic name, and put that name here
NTFY_SERVER Your own ntfy server, if you don't use https://ntfy.sh
SMTP_*, DIGEST_TO Email digest (for Gmail, use an app password with smtp.gmail.com:587)
FRONTLINE_LIBRARY_DIR Default library folder (picking one in the app overrides it). E.g. ~/Documents/Research Papers, or a folder inside Google Drive / Dropbox to sync it
SEMANTIC_SCHOLAR_API_KEY, OPENALEX_MAILTO Optional. Raise API rate limits
PUBLIC_URL Base URL used for links in notifications (default http://localhost:8000). The desktop app sets it for you
FRONTLINE_DB Where the SQLite database lives (default data/frontline.db)

πŸ’Έ What does it cost? Each new paper costs one short AI call (at low effort with Claude). That covers Radar papers and new PDFs in your library. If your library folder already holds hundreds of PDFs, the first scan tags each one once. Deep dives, Q&A and the idea assistant only run when you click them.

Desktop app notes
  • The app opens in its own window, with no browser tab or terminal; the backend runs inside it. Only one window can be open at a time. Logs go to data/frontline.log.
  • macOS: the first time, macOS may ask you to allow notifications from Script Editor, which is what sends them. The app is a small launcher that points at this folder, so if you move the project or recreate .venv, run ./install-desktop.sh again. To uninstall, delete ~/Applications/Frontline.app; your papers and settings stay in data/. Don't run the .ps1 scripts on a Mac: they're for Windows.
  • Windows: if you'd rather use a browser, .\run.ps1 serves the app at http://localhost:8000.
  • The app serves itself on port 8000 when it's free, and on a random port otherwise. Links in ntfy and email notifications only open the app when it's on 8000, so don't leave a separate uvicorn server running on that port.
  • In browser mode, native desktop notifications aren't available. Browser, ntfy and email notifications all work.

Under the hood

Project layout
File Role
frontline/sources.py Paper fetchers (topic keyword search and per-author) and de-duplication
frontline/researchers.py Researcher name search, profiles, arXiv category learning
frontline/summarizer.py AI prompts (cards, deep dive, Q&A, digest) plus the no-key fallback
frontline/sweep.py Orchestration, progress state, scheduler
frontline/notify.py Desktop notifications (Windows/macOS), ntfy and email
frontline/library.py Paper library: folder scanning, PDF metadata, interest tagging, file actions
frontline/backup.py Back up everything to one .frontline file and restore it on another computer
frontline/conferences.py Conference catalog, tracked venues, time zones, calendar export
frontline/planner.py Ideas, idea assistant, projects, tasks and personal deadlines
frontline/db.py SQLite storage
frontline/app.py FastAPI routes
frontline/desktop.py Desktop window (pywebview), Windows shortcuts and macOS .app installer
static/ The single-page UI (no build step): app.js (shared helpers + Radar), home.js, library.js, conferences.js, ideas.js, projects.js

License

MIT. Use it, fork it, make it yours.

About

Frontline stands at the front line of research: it takes the daily barrage of new papers, so you only get the ones that matter, first.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages