About
MOSAIC (Multi-source Scientific Article Indexer and Collector) is a command-line tool that searches 22 scientific databases with one query, keeps every result in a local SQLite cache, and downloads the PDFs you are entitled to read. On top of that library it exports to BibTeX, RIS, Zotero and Obsidian, follows a field with saved searches, and lets a language model rank, summarise, compare and answer questions from your papers.
What it does
| You want to | mosaic | Start with |
|---|---|---|
| find papers on a topic in every database at once | search merges arXiv, OpenAlex, Crossref, PubMed, HAL, Semantic Scholar and 16 more | Tutorial 1 |
| narrow by year, author, journal, title, open access | filters applied by each source and again on every result | Tutorial 2 |
| get the PDFs | the source's link, every open copy Unpaywall knows, your saved logins, your own browser | Tutorial 4, 5 |
| put them in your tools | BibTeX, RIS, CSV, JSON, Markdown, Zotero, Obsidian, NotebookLM | Tutorial 6 |
| hear about new papers | saved searches that report what is new | Tutorial 7 |
| read less | relevance ranking, one-paragraph summaries, comparison tables | Tutorial 8 |
| ask your library | answers with citations from your papers, on your machine | Tutorial 9 |
Principles
- Legal access only. mosaic downloads open-access copies and what your own subscriptions open. It does not get past paywalls, access controls or bot detection; where a publisher blocks automation, it hands the download to your browser.
- Your data stays local. The cache, the PDFs and the vector index live in your home directory. A language model runs where you point it: on your machine with Ollama, or a cloud API you choose.
- Few dependencies. The core needs httpx, typer, rich, tomli-w and rank-bm25; heavier features are extras.
- Every source is a small class. Adding a database means one module; adding a JSON API needs a few lines in the config file (Custom sources).
Where to start
New to mosaic: Installation, then the Tutorial. Looking for one task: the Cookbook. Looking up an option: the Reference.
Authors
Chief Yak Shaver & Accidental Package Maintainer — Fortran programmer who needed one paper, opened 21 browser tabs, and six months later found himself maintaining a Python library
Grand Pixel Overlord & Architect of the Sacred Button — world-class web UI designer, responsible for making MOSAIC actually look good
Claude (Anthropic)
Omniscient Code Oracle & Tireless Rubber Duck — AI pair programmer, responsible for writing the boring parts so humans don't have to
License
MOSAIC is free software, under your choice of GPL-3.0-or-later, BSD-2-Clause, BSD-3-Clause or MIT.