inpi-patents

Where things stand

Credentials (names only, values never leave the machine)

Data directory

The path from licence to table

Request FTP access from INPIFill the database selection form, email it to INPI's licence desk; access codes arrive by email. See the docs: Getting access.

manual

Put the access codes in .envinpi env init writes the template. One USER/PASS pair per licensed database.

once

Map the server, download the snapshotFTP tab: probe, then fetch /Backlog. Resumable, size-verified.

FTP tab

Parse, merge, regionaliseBuild tab: archives to parquet parts, parts to tables, tables to inventors_geo.

Build tab

Explore and useExplore tab: coverage by year, top départements, single-document lookup. The parquet files are the interface.

Explore

Map the server (read-only)

credentials start directory depth

Download archives

Files already complete on disk are skipped; interrupted transfers resume. Re-run any time.

credentials server directory file pattern (regex) destination (optional)

1 · Parse archives to parquet parts

archives (dir or glob; default: everything under data/raw/ftp) workers limit

2 · Merge parts into tables

One parquet per table under data/export; the newest version of each publication wins.

3 · Build inventors_geo

One row per document × inventor with département and NUTS-3, city/postcode swaps repaired.

Explore the export

Build the tables first.

Look up a publication

publication number

Paste an ST36 document

Runs the same parser the pipeline uses, on this machine only. Handy to see how a record is read before you build anything.

merge liste/notice variants