MOSAICOne search, two dozen scientific databases

Search arXiv, OpenAlex, Crossref, PubMed, HAL, Semantic Scholar and 16 more at once, keep every result in a local library, download the PDFs you are entitled to read, and ask a language model about them, on your machine.

Tutorial

Install

Cookbook

Reference

View on GitHub

🔍

Every database in one query

22 sources searched in parallel, from arXiv and OpenAlex to PubMed, IEEE Xplore and Scopus. Duplicates merged by DOI; filters by year, author, journal, title or abstract, and the source's own syntax when you need it.

A first search

📄

The PDFs you can read

The source's link, then every open copy Unpaywall knows, then your saved publisher logins. Each file is checked to be a real PDF and named after the paper.

Getting the PDFs

🔐

Paywalls, legally

Where a publisher blocks automation, mosaic opens the article in your own browser and files the PDF you download. PDFs you already have are imported by their DOI.

Paywalled papers

🗄️

A library that grows

Every result goes into a local SQLite cache that fills in what one source leaves out. Search it offline; export it to BibTeX, RIS, CSV, JSON, Zotero or Obsidian.

The cache

📡

Following a field

Saved searches report only the papers they have not shown you before; one cron line makes a morning digest. Related papers from any seed paper.

Following a field

🧠

A model that reads with you

Relevance ranking, one-paragraph summaries that cite the rows, comparison tables, and questions answered from your papers with citations. Ollama on your machine, or a cloud API.

A local model

🕸️

The citation map

Citation links between the papers of your library, hub papers, clusters, and graph files for Gephi, Graphviz or Mermaid.

The citation network

🤖

Scripts and agents

JSON output for jq and pipelines, exit codes that report failures, a Claude Code skill, and a web interface for colleagues who prefer a browser.

Scripts and agents

One query, every source

The sources answer in parallel; mosaic merges what they return into one table and keeps it.

mosaic search: a spinner while the sources answer, then one merged table of results

Quick start

bash

pipx install 'mosaic-search[all]'
mosaic config --unpaywall-email you@example.org

mosaic search "hall thruster particle-in-cell" --oa-only --download
mosaic get 10.3389/fphy.2019.00140
mosaic search "hall thruster" --cached --output refs.bib

Then the tutorial: ten chapters, one literature search, real output.

A paywalled paper, through your own browser

The terminal side of mosaic get --handoff: the article opens in your browser, you click Download, and mosaic files the PDF under the paper's name.

mosaic get --handoff: the article opens in the browser, the downloaded PDF is moved into the library

MOSAICOne search, two dozen scientific databases

Every database in one query

The PDFs you can read

Paywalls, legally

A library that grows

Following a field

A model that reads with you

The citation map

Scripts and agents

One query, every source ​

Quick start ​

A paywalled paper, through your own browser ​

A paragraph instead of thirty abstracts ​

One query, every source

Quick start

A paywalled paper, through your own browser

A paragraph instead of thirty abstracts