Literature Radar: Self-Hosted AI Literature Monitoring

How It Works

The idea is simple: you should read the best new papers in your field, not skim hundreds of abstracts to find them. Literature Radar splits that job into four steps.

First, you describe your research interests in plain language. This is not a keyword list, it is a short paragraph, the way you would explain your work to a colleague. You also add topic search queries that tell the poller what to pull from each source.

Second, a scheduled job polls the sources on your schedule and collects new papers into a local SQLite database. Nothing is duplicated, papers already seen are skipped.

Third, your AI model reads each new title and abstract and scores it from 0 to 10 against your interest profile, attaching a short reason. A paper about your exact method in a neighboring field can score high, while a paper that merely shares your keywords scores low. This is the part keyword alerts get wrong, and the reason an AI ranking step earns its place.

Fourth, on your chosen schedule, you get one email with the best papers, best first, each with its score and the one line reason it surfaced. Nothing repeats across digests, and everything stays browsable in the admin area afterward.

The Sources It Watches

Literature Radar covers the major open indexes of the scientific literature:

arXiv for physics, mathematics, computer science, and quantitative biology preprints. PubMed for biomedical and life sciences, and adding a free NCBI API key in the config raises the polling rate limit. Crossref for the rest of the journal world, covering nearly every publisher with a DOI. bioRxiv and medRxiv preprints, matched locally against your topics.

Polling is polite by design. Requests identify themselves with your contact email, arXiv queries are spaced out, and the poll window and frequency are yours to set. You are a good citizen of these free services out of the box.

Bring Your Own AI Model

Ranking runs through whichever of these you have, checked in this order:

1. The Claude command line tool, so an existing Claude subscription covers ranking with no per-token cost. 2. The Codex command line tool, the same idea for a ChatGPT subscription. 3. An Anthropic API key. 4. Any OpenAI compatible API, which includes OpenAI itself, OpenRouter, and local servers like Ollama or LM Studio.

There is no Literature Radar account and no third party service in the middle. Your abstracts go to the model you chose and nowhere else.

Setup

Clone the repository, then copy the sample config:

cp config.sample.php config.php

Fill in an admin password, your digest email, and SMTP details, then either serve the public directory with PHP or use Docker:

docker compose up -d

Add the cron job that drives everything:

0 */2 * * *  php /path/to/Literature-Radar/cli/cron.php

Each run polls when a poll is due, ranks anything new, and sends the digest when it is due. You can also run any single step by hand:

php cli/cron.php poll
php cli/cron.php rank
php cli/cron.php digest

The Admin Area

The admin page gives you the ranked paper list with score and source filters, topic management, the interest profile editor, a live digest preview, status counts, and one click poll, rank, and send buttons. It is the place to tune your profile: if the digest is surfacing the wrong papers, edit the profile text, re-rank, and preview the result before the next email goes out.

Who It Is For

Literature Radar fits anyone who needs to track a moving research front: graduate students building a literature review, principal investigators watching for competing results, clinicians following a treatment area, and industry scientists tracking techniques as they mature. It also works for a lab as a whole, one install can track several topics, and the digest email can go to a mailing list.

If you have ever set up journal table-of-contents alerts and watched them bury the one relevant paper under forty irrelevant ones, this is the fix: the same coverage, filtered by something that actually understands what you work on.

Get the Code

Literature Radar is free, open source, and MIT licensed, built by Paul Crinigan at AI Apps API. The source, issue tracker, and README live at github.com/AIAppsAPI/Literature-Radar. If a source you need is missing or a feed misbehaves for your field, open an issue.

It is part of our growing collection of free AI science tools, alongside labparse for lab instrument data and statsage for statistical analysis.