Skip to content
TakoAPI
Back to skills

arxivkb

by camopel

DevOps & Cloud
395Downloads reported by clawskills.sh
0Views
0 likesSign in to like this skill
—Not yet rated
Your rating

Brief

Local arXiv paper manager with semantic search.

Description

What This Skill Does

ArXivKB crawls arXiv's public API by category, downloads PDFs, and indexes their full text using FAISS vector search with Ollama embeddings. Semantic search runs against chunked PDF content rather than just abstracts, and all data stays on-device in SQLite and FAISS files.

Indexes full PDF text in 500-token chunks rather than just abstracts, so searches find papers by what they actually discuss, not only how they were titled or tagged.

When to Use It

  • Tracking new ML papers published in specific arXiv categories each week
  • Searching paper content by concept when you can't recall the title or authors
  • Building a private research library without uploading PDFs to any cloud service
  • Reviewing all papers ingested on a topic over a custom date range
  • Checking how many papers and chunks are indexed across enabled categories

Example Workflow

INPUT

User asks: Find recent papers about contrastive learning in vision transformers

AGENT
  1. 1

    Run /akb add cs.CV cs.LG to enable relevant categories

  2. 2

    Run akb ingest --days 14 to crawl arXiv, download PDFs, and embed chunks

  3. 3

    Query the FAISS index with the natural language phrase

  4. 4

    Retrieve top-scoring chunks grouped by paper

  5. 5

    Return paper titles, abstracts, and matching text excerpts

OUTPUT

A ranked list of papers whose PDF content matches the query, with relevant excerpts showing where each paper discusses the topic

Requirements

  • Ollama installed and running locally (`ollama serve`)

Installation

clawhub install camopel/arxivkb

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links