PatentsView

PatentsView is a data platform from the US Patent and Trademark Office (USPTO) that turns the sprawling US patent record into something you can actually query. Underneath sits a relational database that connects each patent to the people who invented it, the organisations that own it (its assignees), the places involved, and the technology classifications that describe it. The title and abstract of every patent are held as full text, and the citation links between patents are included too, so you can trace how ideas build on one another.

You can reach the data two ways. The Data Download publishes the entire collection as flat CSV tables you load into a database, join, and text-mine offline, which the maintainers recommend for anything large. The API returns JSON for lighter, repeated lookups: it caps requests at 45 per minute per user and lets you ask for exactly the fields you want. A common RAG recipe is to take the bulk tables, join each patent's text to its inventor, assignee, and classification metadata, then chunk each record into title and abstract passages for embedding.

It shines for a prior-art or patent-search assistant, competitive-intelligence tooling, or any system that has to reason about who invented what, where, and when. The structured links are the payoff: you can filter or rerank results by assignee, filing year, or technology class rather than leaning on text similarity alone.

Keep two limits in mind. PatentsView covers granted US patents and pre-grant publications, not filings from other jurisdictions, and the ready-made text stops at titles and abstracts, so full claims and specifications still mean pulling documents from USPTO bulk data. The upside is a clean licence: CC BY 4.0 permits commercial use provided you credit PatentsView.

When you need worldwide coverage or complete patent bodies, pair it with the Google Patents Public Datasets on BigQuery. For well-linked US data, PatentsView is the tidier starting point.