AI Papers Tracker (arXiv + Papers With Code)

Track new AI / agent / LLM research papers from arXiv + Papers With Code, filtered by your own keyword list. Returns a ranked, daily-refreshable dataset of the most relevant recent work.

What does AI Papers Tracker do?

The AI / agent / LLM research output now runs at hundreds of new arXiv preprints per week, and filtering signal from noise is impossible by hand. This Actor tracks two of the highest-signal feeds — arXiv (cs.AI / cs.CL / cs.LG / cs.MA / cs.SE) and Papers With Code (papers that ship runnable code) — filters by your own keyword list (e.g. llm agent, agent verification, code generation evaluation), and returns a deduplicated, scored, ranked dataset.

You get the latest 30-day-window papers ranked by trending score: recency + how many of your keywords matched + relevant arXiv category + whether code is attached. Pair it with the Apify scheduler to run it daily and you have a personal arXiv digest that does not miss anything in your niche. Apify gives you free scheduling, REST API, webhooks, and integrations (Slack / Make / Zapier) on top.

Why use AI Papers Tracker?

• AI researchers: a personalized "what's new in my subfield this week" digest, no manual scraping
• Journalists & analysts: discover papers worth covering before they hit Twitter
• Founders & operators: track benchmarks (agent coding, verification, eval) so your product does not lag the frontier
• Investors: map which labs / authors keep publishing in a thesis area
• Course / curriculum builders: weekly refresh of recommended reading

How to use AI Papers Tracker

1. Open the Input tab
2. Edit the Keywords list — phrases that should appear in titles / abstracts (e.g. RAG evaluation, tool-use agent, code review LLM)
3. Set Days back (default 30) and Max results (default 30)
4. Click Start
5. Download the Dataset or the human-readable ai-papers-report.md from the Storage tab
6. Schedule it daily / weekly in the Schedules tab for a continuous feed

Input

• keywords — array of phrases. Each phrase is searched independently across arXiv title / abstract + PWC search. Results are deduped by arXiv ID.
• daysBack — only include papers submitted in the last N days (1-365)
• maxResults — cap the dataset to the top N scored papers (1-200)

Output

Each row in the dataset:

{
  "title": "MAST: Multi-Agent System Failure Modes",
  "authors": ["Cemri et al."],
  "abstract": "First 400 chars of the abstract…",
  "arxivId": "2503.13657",
  "url": "https://arxiv.org/abs/2503.13657",
  "submittedAt": "2026-05-15",
  "categories": ["cs.AI", "cs.MA"],
  "source": "arxiv",
  "hasCode": true,
  "matchedKeywords": ["agent verification", "llm agent"],
  "trendingScore": 78
}

You can download the dataset in various formats such as JSON, HTML, CSV, or Excel.

Data table

Field	Type	Description
title	string	Paper title
authors	string[]	Authors list (full for arXiv, first few for PWC)
abstract	string	Truncated to 400 chars
arxivId	string | null	arXiv identifier (e.g. 2503.13657) when available
url	string	Direct link to the paper page
submittedAt	date	YYYY-MM-DD submission date
categories	string[]	arXiv categories (e.g. cs.AI)
source	string	arxiv or paperswithcode
hasCode	boolean	Paper appears on Papers With Code (i.e. ships code)
matchedKeywords	string[]	Which of your keywords matched this paper
trendingScore	number	0-100 composite ranking score

Trending score (max 100)

Dimension	Max	What it measures
Recency	50	Closer to today within the window = higher
Keyword relevance	32	Number of distinct keywords matched (8 each)
Category fit	10	Counts cs.AI / cs.CL / cs.LG / cs.MA / cs.SE
Code attached	8	Paper appears on Papers With Code

Pricing / Cost estimation

Pay-per-result: roughly $0.05 per 10 papers scored. A daily refresh tracking 4 keywords over 30 days typically returns 30-60 results = under $1/day. The Apify free trial covers your first run end-to-end so you can validate output before subscribing.

Tips / Advanced options

• Cluster keywords semantically: do not add both LLM and large language model — they will overlap ~80% and double your runtime.
• arXiv rate limit: this Actor sleeps 3.1 seconds between arXiv calls (arXiv's published rate-limit courtesy delay). With 8+ keywords expect a ~30s run.
• Schedule weekly + diff: for a real "what's new this week" digest, schedule weekly and diff the dataset against last week's run.
• Combine with mcp-server-catalog: this Actor catalogues the research; that one catalogues the tools. Together = full landscape view.

FAQ, disclaimers, and support

Why arXiv + Papers With Code specifically?

arXiv = full firehose of preprints (titles + abstracts indexed). Papers With Code = the subset that ships code (a strong quality + reproducibility signal). Together they cover ~99% of what an AI engineer or researcher actually needs to read.

Is the data current?

Both APIs return live data at run time. Schedule daily for a true real-time feed.

Why is paper X not in the results?

Either (a) it was submitted outside the daysBack window, (b) none of your keywords matched the title or abstract, or (c) it is published in a non-cs.* arXiv category (e.g. stat.ML — open an issue to request).

Legality

The arXiv API and Papers With Code REST API are both public, documented APIs with no rate limit beyond the arXiv 3-second courtesy delay (which this Actor respects). No login, no scraping of restricted content.

Support

Open an issue on the Issues tab of this Actor's Apify Console page, or via the author's GitHub.

Built by Ian Mu (github.com/ianymu) — author of verify-before-stop, a Claude Code harness that gates session-stop on real, verified output.