WINNOWR

What WINNOWR is

WINNOWR is a CPU-first engine for identifying potential web-source matches in student writing and other long-form text. Its reports provide index overlap, candidate source URLs, and highlighted spans for review. Matching uses deterministic text normalization and compact word-shingle fingerprints to surface candidate documents. Institutions can create an index from their own JSONL.zst corpus, and the repository documents workflows using FineWeb-Edu and Wikipedia. The HTTP API accepts text or file submissions and includes endpoints for submissions, reports, text, purging, and health checks. The documented runtime uses SQLite, Cloudflare Queues, Cloudflare R2, and a local sharded index on disk. Build and deployment instructions cover the API and worker, including plain Linux machines; the project describes an index builder, API server, and background worker. Longsight presents WINNOWR as academic-integrity software for institutions, focused on plagiarism detection and reviewable evidence. The project is usable but still evolving. It is licensed under GNU AGPLv3, whose network-use provision requires modified deployed versions to make those modifications available under the same license. Source-page snippets and highlighted spans may be unavailable if a live page cannot be fetched or parsed.