# Papers We Love: 54 topic folders of links, and one Shell script that fetches what they point at

> Papers We Love is a GitHub directory of computer science papers kept as markdown links spread across 54 topic folders, not as hosted PDFs. The curation is the asset; the hosting is deliberately somebody else's problem, and one Shell script walks the links to pull down what it can.

**papers-we-love/papers-we-love** — Papers from the computer science community to read and discuss.

- Repository: https://github.com/papers-we-love/papers-we-love
- Website: http://paperswelove.org/
- Stars: 110,108 · Forks: 6,442
- Language: Shell
- License: not declared
- Published: 2026-08-17 · Updated: 2026-08-18 · Language: en
- Canonical page: https://hysenlabs.com/projects/papers-we-love-papers-we-love

## Every entry is a URL, and licences decide which ones carry a file

The README gives the reason for the shape of the project in one sentence: because of licences, the project cannot always host the papers themselves, so it links to wherever a paper lives instead. The exception is marked, with a scroll symbol beside the title in the directory, and the contributing guide keeps a section on respecting content licences for the cases that come up when someone wants a PDF added to a topic folder.

Consequence for the reader: a clone of papers-we-love/papers-we-love is an index, not a library. You get a directory of pointers, and every pointed-to file is hosted by an arXiv mirror, a university, or an author, each of which can move, paywall, or vanish. Nothing in a checkout tells you which of those links still resolve, and no job in the repository checks them for you. Reading from this list means trusting a third party's copy as much as the curation that pointed at it.

## The top level is 54 topic folders and one file nobody explains

Look at the top level and the shape of the project is immediately visible: 54 directories named by field, plus .github/, .gitignore, AGENTS.md, CODE_OF_CONDUCT.md, README.md, and a single file called nautilus.db. The topic directories are the entire navigation, running from affective_computing/ and audio_comp_sci/ through crash_only/, data_fusion/, gossip/, languages-theory/, memory_management/, non_blocking_algorithms/ and pattern_matching/.

Two details of that layout cost the reader time. There is no top-level index file, so there is nothing to search across the repository without pulling it down and grepping the markdown yourself. And the naming is inconsistent: some topics use underscores such as data_structures/, others use hyphens such as brain-computer-interface/ and distributed-file-systems/, so a script that globs by convention silently misses entries. Consequence: nautilus.db carries no description anywhere in the files the repository exposes, and its name gives no hint whether it is an index, a cache, or a leftover.

## download.sh reads the markdown, finds PDF links, and writes into the topic folders

The repository's primary language is Shell, and the shell is one script. The README gives the command and describes what it does: scrape the markdown files for links to PDFs, then download the papers into their respective directories.

```bash
git clone https://github.com/papers-we-love/papers-we-love.git
cd papers-we-love
./scripts/download.sh
```

Additional options are kept in scripts/README.md rather than in the top-level README. Consequence for the reader: the script operates on links it finds in text, so an entry that cites an abstract page, a DOI, a thesis behind an institutional login, or a personal home page is not a paper it can fetch. The description mentions no list of failures and no summary at the end of a run, which means a download that retrieved 40 papers out of 60 entries looks exactly like one that retrieved all 60. Count the files yourself before treating a topic folder as complete.

## The list of other places to read is the honest part of the README

Papers We Love does not pretend to replace arXiv, and the section on other places to find papers runs to roughly two dozen entries: arXiv itself, Lobste.rs filtered to PDF, SciRate, cat-v.org, the Bell System Technical Journal for the years 1922 to 1983, hand-built bibliographies like the Gradual Typing Bibliography and the Services Engineering Reading List, a Microsoft Research publications page, and The Morning Paper blog.

The alphaXiv entry states the difference in approach in a single sentence: it adds a discussion layer, and the way to use it is to replace arxiv with alphaxiv in an arXiv paper URL. Same identifier, different front end, no download step. Consequence for the reader: if you want the PDF of a paper you can already name, this repository is a detour that adds a hop. If you want to know what a reading group thought was worth an evening on a subject, the topic folders carry a selection no search index reproduces, and the value sits in the choosing rather than in the hosting.

## Chapters are the part that actually meets people, and the logo is not yours

The README points at local chapter meetups, a Discord server for events and for discussion of the repository's contents, a YouTube channel holding past presentations as videos and playlists, and a separate organizers repository for people who want to start a chapter in their city. Every meetup runs under CODE_OF_CONDUCT.md, which sits in the repository next to the topic folders rather than on the site.

The Copyright paragraph is short and specific about the constraint. The name Papers We Love and the organization logos are copyrighted, owned by Papers We Love Ltd, all rights reserved, and starting a chapter means reading the chapter-creation guidelines and asking before using the logo. Consequence: the directory of links can be copied and its curation used freely enough, but a fork cannot present itself as a Papers We Love chapter, and the trademark stays attached to the meetup brand rather than travelling with the list of papers.

## A contribution is a markdown patch, with no published schema behind it

The README names exactly which pull requests it wants: papers the project should add, better organization of the papers it has, and links to other paper repositories worth pointing at. The contributing guide lives in CONTRIBUTING.md, and the licence question raised by the hosted-or-linked decision comes back there.

Consequence for the reader: adding a paper is a patch to a text file, which is a low bar, but no review checklist is published for the curation decision itself, and there is no per-entry schema. A useful addition also has to be findable, and with 54 undifferentiated topic directories and no index, a paper filed under the wrong one is invisible to the people who would have wanted it. If you maintain a fork, you are maintaining a diff against a repository with no releases and no changelog, so the commit stream is the only record of what the list now contains.

## No licence file at the top level, and no release to pin

Two facts from the repository's own top level decide how much weight this list can carry. The first is the missing licence: the entries are .github/, .gitignore, AGENTS.md, CODE_OF_CONDUCT.md, README.md and nautilus.db, GitHub identifies no licence for the project, and the Copyright paragraph covers the name and the logos only, saying nothing about the directory of links itself. Consequence: a reader who needs to know the terms the curated list is offered under cannot get them from the repository.

The second is the absence of releases. The last push to main was on 2026-09-29, so the topic folders are being edited, but the project publishes no GitHub releases, which means there is no version to cite, no changelog to diff, and no marker telling you whether the topic folder you read last year holds the same set of papers it does now. The only record of additions is the pull request stream, and the only fix for a stale, dead, or misfiled entry is another pull request.

## Conclusion

Papers We Love earns a place in a reading queue if you want a human selection organised by field and you are willing to follow links to wherever a paper is hosted, and the chapter network is the part worth joining if the meetup format is what you came for. It is the wrong tool when you need the PDF in one place, when you need to know a link still resolves, or when you need to cite a stable version of the directory, because nothing in the repository states a licence for its content and nothing is versioned. Before you build anything on it, check three things: which topic folder would hold your subject, whether the papers you care about are linked or hosted, and what the actual hosting terms are for the content you copy out of it.

## FAQ

### How can I find a paper in the Papers We Love directory?

Clone the repository and look inside the topic folders, which are named by field, such as distributed_systems/, gossip/ or information_theory/. There is no top-level index file and no search, so you either guess the folder or grep the markdown files for the title you have in mind.

### Where can I find research papers on GitHub through Papers We Love?

The repository stores each paper as a link inside a topic folder rather than hosting the file, because of content licences. A scroll symbol next to a title in the directory means that one is hosted there; everything else points to an arXiv mirror, a university page, or the author's own site.

### What does scripts/download.sh do in Papers We Love, and what does it miss?

It scrapes the markdown files for links to PDFs and downloads the papers into their respective directories, with more options in scripts/README.md. It fetches only what the links point at, so entries citing an abstract page, a DOI, or a paywalled thesis are not downloaded, and the README describes no failure report for a run that comes up short.

## Sources

- [Official documentation](http://paperswelove.org/)
- [Official README](https://github.com/papers-we-love/papers-we-love#readme)
- [Project repository](https://github.com/papers-we-love/papers-we-love)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/papers-we-love-papers-we-love