citesync

Software citation metadata.

On this page

Generate citation files in the browser

Open the playground to generate citation files from one metadata file or a form with project details and authors. You do not need Python installed. Your files stay in browser memory; they are not uploaded or saved between visits. Downloading the converter requires a network connection.

Paste a file or enter project details

Paste or open CITATION.cff, codemeta.json, .zenodo.json, pyproject.toml, Cargo.toml, package.json, PKG-INFO, METADATA, R DESCRIPTION or Julia Project.toml/JuliaProject.toml. Pasted content is detected automatically; use the filename selector to override an uncertain result. Both file selectors group their files and sort them by name, and you can type in them to filter, by filename or by group (python, bibtex). An opened file with a recognised filename uses that filename. Other text files are detected from their contents. Try an example follows the selected format; in automatic mode it uses the recognised format, or pyproject.toml for an empty or ambiguous input. While the example is unchanged, choosing another input format switches the example too. Editing or opening a real file ends that behaviour. Each request reads only that file, ignores its embedded citesync settings and uses a temporary repository in browser memory. It does not execute package code or fetch metadata from repository links. Inputs are limited to 1 MB.

Switching entry methods transfers project metadata. An unchanged file draft keeps its original text and format. Editing details creates a CodeMeta working copy for the file view, retaining extra CodeMeta fields, author identifiers, affiliations and Role objects. Other formats are projected through the engine: formatting and unmapped fields from CFF or package manifests are not represented in the CodeMeta form. Keep the original file when those matter. Partial drafts transfer too. Invalid file syntax cannot populate the form; the original file and any previous details remain available, with a notice. Failed or superseded transfers cannot overwrite a newer edit. Unchanged uploaded CFF files retain their original line endings until edited.

Review author names

The form accepts multiple authors with explicit given and family names. Imported one-string names can be edited in an optional Author names disclosure. It opens for potentially ambiguous cases: single or multiple words, initials, punctuation, and East Asian scripts. Names look right, or closing the disclosure, keeps it closed until the imported names change. Corrections stay while you edit the source, as long as the imported names are the same. Two-word names can still have a different order; this is a prompt, not a confidence score. Downloads do not require a confirmation checkbox. Corrections update the preview automatically. Editing the source or switching entry methods disables old downloads immediately, then refreshes after a 100 ms typing pause. Updates keep keyboard focus in the field being edited. Entry-method tabs use arrow keys to move focus and Enter or Space to open a view; snippet tabs update immediately. Drafts remain on the page after an error.

Copy or download the results

The playground creates CITATION.cff, codemeta.json and .zenodo.json. Existing metadata is updated where the engine supports it. Conversion notes report fields a destination cannot carry. The preview preserves source metadata where the engine does, including CodeMeta Role objects; it does not invent unavailable values. Zenodo output is metadata for review, not a submission to Zenodo.

BibTeX (citation.bib, a biblatex-software @software entry), CSL-JSON (citation.csl.json, type software), an APA 7 reference (citation.apa.txt), schema.org JSON-LD for a website (schema.jsonld), DataCite XML (citation.datacite.xml, only when the record has a DOI, an author, a publisher and a release year), RIS (citation.ris) and EndNote (citation.enw) are also available, the same files citesync export prints. They use the reviewed author names. A release year/date is included only when supplied and readable. They are output formats, not accepted input formats. CSL-JSON follows the official CSL input schema.

The output selector groups citation files and Package templates (incomplete) together. Citation choices include short destination labels: GitHub, software metadata, Zenodo, BibTeX, CSL-JSON, APA reference, website structured data, DataCite metadata, and RIS and EndNote references. Templates are labelled incomplete in the selector, with their limits below the preview actions. It contains metadata-only pyproject.toml, Cargo.toml, package.json, R DESCRIPTION, Julia Project.toml, Typst typst.toml and Quarto _extension.yml, each labelled with its language, not complete package configurations. The DESCRIPTION splits a Package: Title name into its two fields and writes Authors@R with the aut and ctb roles; the Project.toml writes authors as CFF tables and leaves the uuid for Pkg to make; the typst.toml writes a [package] table with a major.minor.patch version and no ORCID, and the _extension.yml writes a title, a version and one author string. Add build settings, source files and dependencies before use. Names and versions that cannot be carried over are omitted with notes; no identifier or version is invented. The npm template is marked private, and only its first author is included because npm has one author slot. Copy file and Download file sit above the preview, with copy as the primary action. The secondary citation ZIP is generated only on request, avoiding archive work during previews. Templates download individually; there is no template ZIP.

The ZIP preserves filenames, including the first dot in .zenodo.json. Some browsers strip that dot from an individual download; rename the file or use the ZIP. Clipboard access depends on browser permission.

Conversion notes show omitted values, warnings and errors, with remedies where available. Missing optional fields are not shown as problems. Template notes relate to the selected template.

Field examples remain visible while typing. Version and SPDX help uses native disclosures and official reference links, accessible by keyboard and touch. Validation guidance shares the amber note style; copying and loading keep neutral feedback.

Look up a field

Open Field lookup to search CodeMeta terms and field paths across CITATION.cff, codemeta.json, .zenodo.json, pyproject.toml, Cargo.toml and package.json. Each format section shows the pinned codemeta-drift mapping, including fields with no representation. The source link identifies the pinned release and commit. Lookup works independently of the Python converter. A mapped path does not mean citesync writes that field; see supported files for write limits.

Use citesync in a repository

The repository-use section provides copyable CLI commands, a pre-commit/prek hook and GitHub Actions steps, each with a labelled copy icon in its code block. The checks require committed citation files and fail on disagreement; they do not rewrite files.

Run locally

uv sync --locked
pnpm install --frozen-lockfile
just playground

Open http://127.0.0.1:8468. This serves the same site/ bundle that is deployed, with the playground at the root and documentation under docs/.

The playground uses the Python engine from this checkout through Pyodide 314.0.7 in a worker. Python dependencies come from uv.lock, with artifact hashes checked during the build and again when loading. There is no separate JavaScript converter. The page prepares Pyodide from jsDelivr in the background, downloading converter packages concurrently. Loading feedback follows download, initialization and package setup stages; it does not estimate completion time. Offline operation is not promised.

Check

just check
just playground-build
pnpm exec playwright install chromium firefox webkit
pnpm run test:browser

Browser tests run the same journeys in Chromium, Firefox and WebKit using Playwright’s pinned browsers. Set CHROME_PATH to override only Chromium. CI installs browser system dependencies; local machines need the libraries reported by Playwright. They run actual Pyodide conversion, name correction, form generation, parse-error recovery, runtime-download failure/retry, stale-export protection, typing during startup, ambiguous detection, focus retention, template downloads and accessibility checks. Network assertions check that metadata is not included in requests and drafts are not saved to web storage. Native copy actions run in all three engines. Clipboard contents are also read back in Chromium, which supports Playwright’s clipboard permission grants. Screenshots go in test-results/. R imports accept only the documented static Authors@R subset; uploaded R is never evaluated. Julia UUIDs are not DOIs. See supported files for role and licence limits. No physical-device or guaranteed offline test is included.

Publish

The playground workflow builds and tests pull requests. On main it deploys the same build to GitHub Pages. In the repository's Settings → Pages, choose GitHub Actions as the build source before the first deployment. PR browser screenshots are available in the playground-browser-results artifact.

Search and sharing metadata are in the initial HTML: a descriptive title, format-specific description, canonical URL and Open Graph/Twitter tags. The canonical deployment URL is https://simonvanlierde.com/citesync/, following the account's GitHub Pages domain. The site is deployed through GitHub Actions with HTTPS enforced; metadata alone does not establish indexing.

The build also writes site/, with content-versioned JavaScript, styles, worker and wheel URLs. This keeps cached assets from different deployments together. The public runtime manifest and wheels use normal HTTP caching; citation drafts are never cached. CI runs the browser tests against that deployment bundle.

The same build renders the Markdown files in docs/ and their linked repository guides under site/docs/. Edit those source files to update the published guides. Documentation and the playground share navigation, typography and colours. Guide links and heading anchors stay on the website. Full-text search uses a same-site index in the browser, independently of the Python converter. On desktop, search and guide navigation sit beside the article. On phones, open Browse documentation to choose a guide. Code examples have copy buttons with feedback if the browser blocks clipboard access. Reading and browsing guides still work without JavaScript.

The page has no upload endpoint, analytics or draft storage. A visitor's files stay in memory and disappear on reload. Runtime downloads contact jsDelivr; project links navigate to GitHub. The local server is for development, not public hosting.