quelle 0.1.1

Fetch academic publication metadata and PDFs from open sources (OpenAlex, Crossref, Semantic Scholar, arXiv, Unpaywall) and return normalised JSON

quelle

quelle is a local CLI that fetches academic publication metadata and PDFs from open academic sources (OpenAlex, Crossref, Semantic Scholar, arXiv, Unpaywall) and returns them as normalised JSON. Designed as a composable building block — feed the output into any note-taking system, reference manager, or research workflow.

The name is German for source — in academic German, "Quelle:" is the word that introduces a bibliographic reference, and fetching from open sources is exactly what the tool does.

What it does

Given a publication identifier or a free-text title, quelle fetch returns a normalised JSON blob with title, authors, year, venue, DOI, abstract, citation count and (optionally) a downloaded local PDF. It walks a fallback chain of free open sources:

Source	Role	Rate limit
OpenAlex	Primary metadata + OA PDF URL	$1/day on free key
Crossref	DOI-authoritative fallback (abstract, journal block)	polite pool (no hard cap)
Semantic Scholar	Citation graph + metadata fallback	5000 / 5 min unauth
arXiv	Preprint metadata + direct PDFs	1 req / 3s (enforced)
Unpaywall	DOI → OA PDF lookup	100k / day

Google Scholar URLs are not supported: Scholar has no public API and its Terms of Service prohibit automated access. If you only have a Scholar link, open the page, copy the paper title, and feed that to quelle fetch as a free-text query — OpenAlex and Crossref cover almost every paper with a DOI.

Stack

Python 3.12+, uv-managed. Typer (CLI) + httpx (sync HTTP) + stdlib sqlite3 (cache) + rich + environs + platformdirs + pytest + pytest-httpx. No GUI, no ORM, no async.

Installation

Install from PyPI:

pipx install quelle
# or: uv tool install quelle

Both install quelle into its own isolated venv and put it on your $PATH. Run the one-time bootstrap to create the config, data, and cache directories and seed a default .env:

quelle init
quelle config edit        # opens the .env in your $EDITOR

Development from a source checkout

git clone https://github.com/vcoeur/quelle.git
cd quelle
make dev-install          # uv sync --all-groups
make test                 # pytest
make lint                 # ruff check + format --check
make format               # ruff --fix + format
uv run quelle --help      # run the CLI straight from the repo

When run from the repo, quelle picks up the .env at the repo root and stores dev cache / PDFs under a repo-local .dev-state/ instead of polluting your installed user data.

Configuration

quelle follows each OS's standard "config dir + data dir + cache dir" layout via platformdirs:

Role	Linux (XDG)	macOS	Windows
Config (.env)	~/.config/quelle/	~/Library/Application Support/quelle/	%APPDATA%\quelle\
Data (downloaded PDFs)	~/.local/share/quelle/	~/Library/Application Support/quelle/	%LOCALAPPDATA%\quelle\
Cache (sqlite index)	~/.cache/quelle/	~/Library/Caches/quelle/	%LOCALAPPDATA%\quelle\Cache\

Any of the three can be overridden via env vars — useful for tests, Docker, or custom deployments:

export QUELLE_CONFIG_DIR=/etc/quelle
export QUELLE_DATA_DIR=/srv/quelle/data
export QUELLE_CACHE_DIR=/var/cache/quelle

Inspect the resolved paths at any time:

quelle config path        # plain output, one path per line
quelle config path --json # JSON, scriptable
quelle config show        # all values including API keys (redacted)

The only variable worth setting by default is QUELLE_CONTACT_EMAIL — it goes into the User-Agent header and enrolls you in the Crossref / OpenAlex polite pool. See .env.example for the full list.

Dev mode: when you run quelle from a source checkout (uv run quelle … inside the repo), the .env at the repo root is still picked up — the same ergonomics as before — but downloaded PDFs and the cache go into a repo-local .dev-state/ directory so your installed user data stays clean.

Migration from PublicationManager: if you upgraded from the old publication-manager package, the first run of quelle automatically moves your ~/.config/publications/.env and ~/.publications/.publications-state/ into the new locations. No data loss, no manual steps.

Usage

# Resolve by DOI (uses OpenAlex + Crossref enrichment by default).
quelle fetch 10.1109/83.902291

# Resolve by arXiv id, with PDF download into the data dir.
quelle fetch 1706.03762 --download-pdf

# Resolve by free-text title.
quelle fetch "The Perceptron: A Probabilistic Model" --json

# Bypass the local cache and force network.
quelle fetch 10.xxxx/yyyy --no-cache

# Inspect the cache.
quelle cache stats
quelle cache list --limit 20
quelle cache show 10.1109/83.902291
quelle cache clear --yes

Claude Code skill

A minimal example SKILL.md ships at the repo root — drop it into ~/.claude/skills/quelle/ (or <project>/.claude/skills/quelle/) to use the CLI from a Claude Code session. It is deliberately thin: resolve the paper, print the metadata, stop. Adapt the last step for your own downstream workflow (Zettelkasten import, BibTeX append, etc.). The skill also includes a prominent reminder about each upstream's Terms of Service and rate limits — read it before running quelle at any non-trivial scale.

Layout

quelle/
  models/        <- Publication, Author (pure dataclasses)
  repositories/
    cache.py           <- SQLite cache keyed by DOI / arXiv / title
    errors.py          <- Error hierarchy -> exit codes 1/2/3/4
    http_client.py     <- httpx + polite User-Agent
    pdf_downloader.py  <- Streaming PDF download with content-type + size checks
    sources/           <- One module per source: openalex, crossref, semantic_scholar, arxiv, unpaywall
  services/
    resolver.py         <- Source orchestration + enrichment chain + cache lookup
    pdf_resolver.py     <- Lazy PDF fallback chain
  cli/
    main.py             <- Typer app (fetch, cache, version, init)
    config.py           <- `config show` / `path` / `edit` + bootstrap
    output.py           <- JSON vs rich TTY rendering
  paths.py              <- platformdirs resolution (config / data / cache)
  migrate.py            <- One-shot migration from the legacy PublicationManager layout
  settings.py           <- environs-layered config
tests/

Layer rules: imports only go downward. Models import nothing from this project. Repositories import models. Services import models + repositories. CLI is the wiring layer.

Status

v0.1 — all five open-API sources wired up with a merge-logic enrichment chain, a SQLite cache keyed by DOI / arXiv id / OpenAlex id / title (second query for the same paper is offline), and a PDF download chain with OpenAlex → arXiv → Unpaywall fallback plus content-type and size validation.

Usage and terms

This tool is intended for personal and academic research use. It queries free, public APIs on your behalf. You are responsible for complying with each upstream's terms of service — the MIT licence on this repo covers the code of this tool, not the data you fetch through it.

Not supported use cases:

• Bulk scraping / batch ingestion of many records. Most upstreams publish free database snapshots; use those instead of hammering the live API.
• Rehosting downloaded PDFs on a public server. The --download-pdf flag writes to a local cache on your machine — that is fine. Re-serving arXiv PDFs, publisher PDFs, or full text from your own infrastructure is not (see arXiv and Semantic Scholar rows below).
• Commercial repackaging of the JSON output as a paid product. Individual commercial use of the metadata is generally allowed by the underlying licences, but Semantic Scholar in particular requires attribution and some S2 records are CC BY-NC.

Per-source summary:

Source	Data licence	Rate limit	Attribution	Notes
OpenAlex	CC0 — "OpenAlex data is and will remain available at no cost"	~100k / day on the polite pool; single-entity lookups unlimited	not required	Provide an email via QUELLE_CONTACT_EMAIL for the polite pool, or set OPENALEX_API_KEY for the new key-based tier (OpenAlex announced in January 2026 that key authentication is replacing the mailto polite pool; the tool supports both).
Crossref REST	CC0 for metadata — "almost none of the metadata is subject to copyright, and you may use it for any purpose". Some abstracts may remain copyrighted.	No hard cap; the polite pool is requested via your mailto= / User-Agent	not required, but recommended	Commercial users who need SLAs should subscribe to Metadata Plus directly with Crossref.
arXiv API	Metadata CC0. PDFs retain their authors' / arXiv's licence.	1 request / 3 seconds (the tool enforces this globally via a module-level lock)	Do not claim arXiv endorses your project.	You may not store and re-serve arXiv e-prints (PDFs, source files, other content) from your own servers unless you have the copyright holder's permission. Downloading for local personal reading is explicitly allowed.
Semantic Scholar	S2 data may be CC BY-NC or ODC-BY depending on the record. The API itself is provided "AS IS, WITH ALL FAULTS, AND AS AVAILABLE" with no warranty.	Public endpoints need no auth; higher throughput requires a free key from Ai2.	Required — "Licensee will include an attribution to 'Semantic Scholar'", and publications must cite The Semantic Scholar Open Data Platform.	You may not "repackage, sell, rent, lease, lend, distribute, or sublicense the API". This tool is a personal client, not a proxy.
Unpaywall	CC0 data	100k requests / day	not required	The email parameter is mandatory — Unpaywall uses it to contact you if something goes wrong. Don't fake it. For bulk workloads, download the free data snapshot instead of hammering the API.

Google Scholar is not supported. Google Scholar has no official API, and Google's Terms of Service prohibit automated access. Passing a Scholar URL to quelle fetch returns a UserError asking you to copy the paper title manually and retry — OpenAlex and Crossref together cover almost every paper with a DOI, so the workaround is usually one extra copy-paste.

No warranty: see the MIT LICENSE — this tool is provided as-is, with no guarantee that its JSON output is correct, complete, or current. Verify critical metadata against the canonical upstream before relying on it.

Licence

MIT — see LICENSE.

Questions or feedback

This is a personal tool — I'm happy to hear from you, but there is no formal support. The best way to reach me is the contact form on vcoeur.com.