You can host your own copy of OpenAlex

Claim

OpenAlex's data is downloadable in full without a key or an AWS account. A paper tool does not need the whole thing: a mirror of the works for its subjects, their references and citers, plus authors and sources, fits in tens of gigabytes and removes the API-credit dependency entirely. The parts of OpenAlex that are hard to rebuild (author disambiguation, institution matching, topic classification) arrive precomputed in the snapshot. The parts you give up are daily updates, which are a paid plan, and the hosted full-text search over 475M works.

Evidence

• Public snapshot in the openalex S3 bucket under data/, downloaded with --no-sign-request; AWS Open Data covers transfer. September 2026 release: about 626M records per format; JSON Lines about 745 GB compressed (works about 615 GB), Parquet about 770 GB; several TB decompressed. Free snapshot refreshes quarterly (second Wednesday of January, April, July, October); paid plans get daily snapshots and change files. Works include abstract_inverted_index (null for some) and the same schema as the API, so referenced_works is present. Partitioned by updated_date with manifests for incremental sync. Filter locally on is_xpac to match the API's default view.
• Licence: the page points to LICENSE.txt at the bucket root rather than stating it. OpenAlex has published under CC0 historically; confirm from the file before redistributing.
• The REST API now bills credits past a $1 per day free budget and requires a key for cursor paging past page one.
• Semantic Scholar's Academic Graph (S2AG) offers bulk dataset downloads of papers, authors, citations, venues and SPECTER2 embeddings (about 214M papers, 2.49B citations); licence is behind an API licence agreement page; API key limit is 1 request per second.

Open questions

• Confirm the licence text in LICENSE.txt.
• Whether S2AG's bulk licence permits storing SPECTER2 vectors in a private index (it is behind the API licence agreement).
• Disk and memory for DuckDB over the works Parquet on VM 104.

Sources

• https://help.openalex.org/access/snapshot — docs, Wed Oct 21
• https://developers.openalex.org/download/overview — docs, Wed Oct 21
• https://help.openalex.org/how-to-use-the-api/rate-limits-and-authentication — docs, Wed Oct 14
• https://www.semanticscholar.org/product/api — docs, Wed Oct 21

AI-assisted research: gathered and summarised with AI from the sources above, each claim checked against them.