rdatagouv is an R client for the public API of data.gouv.fr, the French government’s open data platform. It helps you find a dataset that matches your interests, judge whether it is usable, download it, and later re-fetch the exact same table reproducibly. Requests are built on top of the httr2 package.
Installation
# From GitHub once the package is published
# remotes::install_github("astamm/rdatagouv")Quick start
rdatagouv revolves around a simple workflow: find a dataset, judge whether it is usable, fetch it into a table, and later re-fetch the exact same table reproducibly. The examples below hit the live data.gouv.fr API and show real results.
1. Find
dg_find_datasets() searches the catalog by keyword and returns one row per dataset, with metadata (quality_score, views, license, …) to help you choose:
library(rdatagouv)
hits <- dg_find_datasets(q = "vélo", n = 5)
hits[, c("title", "id", "organization", "quality_score")]
#> # A tibble: 5 × 4
#> title id organization quality_score
#> <chr> <chr> <chr> <dbl>
#> 1 "Statistiques de subventions d’achat de vélo… 63a3… ile-de-fran… 0.889
#> 2 "Fréquentation mesurée dans les Parkings Vél… 63a3… ile-de-fran… 0.889
#> 3 "Nombre de places de stationnement vélo " 67ca… ecolab-1 0.889
#> 4 "Vélib - Vélos et bornes - Disponibilité tem… 5a4e… ville-de-pa… 0.889
#> 5 "Plan Vélo 2021-2026" 6271… ville-de-pa… 0.778You can also discover producers with dg_find_organization() or curated themes with dg_find_topics(), and narrow a search to one of them via the organization / topic arguments.
2. Judge
Before pulling, glimpse a dataset’s health and engagement in one call, or read its documented columns with dg_schema():
g <- dg_glimpse("6a6be5976a05df136d48fb7a")
g$quality$score
#> [1] 0.55555563. Fetch
dg_pull_dataset() downloads the dataset’s first tabular resource and returns a single tibble, with a stable table id attached as an attribute:
tbl <- dg_pull_dataset("6397c0ff56d3963118a18345") # C.vélo bike stations
head(tbl[, 1:6])
#> # A tibble: 6 × 6
#> station_id name physical_configuration lat lon altitude
#> <int> <chr> <chr> <dbl> <dbl> <dbl>
#> 1 4 04 UCA - Campus Cézeaux REGULAR NA NA NA
#> 2 7 07 - Delille REGULAR NA NA NA
#> 3 8 08A - Gaillard REGULAR NA NA NA
#> 4 9 09 - Chamalières Mairie REGULAR NA NA NA
#> 5 14 14 - Les Carmes REGULAR NA NA NA
#> 6 19 19 - Amboise REGULAR NA NA NA
dg_table_id(tbl)
#> [1] "https://www.data.gouv.fr/datasets/6397c0ff56d3963118a18345#01f5b3da-8d58-42c6-a07d-202538ad6672"4. Re-fetch reproducibly
Save that id, and dg_refetch() gets back the exact same table later — no matter how titles or file names drift on the platform:
again <- dg_refetch(tbl)
identical(again, tbl)
#> [1] TRUE5. Summarise
dg_summary() computes metrics for one table (size, columns, rows, missing-value rate), and dg_summarise() does the same for many at once — even an entire dg_find_datasets() result:
dg_summarise(datasets = list(iris = iris, mtcars = mtcars))
#> # A tibble: 2 × 7
#> dataset size_kb n_vars n_numeric n_non_numeric n_rows prop_missing
#> * <chr> <dbl> <int> <int> <int> <int> <dbl>
#> 1 iris 7.09 5 4 1 150 0
#> 2 mtcars 7.04 11 11 0 32 0For a deeper tour — including column-type control and handling parsing problems — see the vignette.
Supported formats
dg_pull_dataset() downloads the first tabular resource of a dataset among CSV, CSV.GZ, XLS, XLSX, PARQUET, TSV, TXT and JSON, and returns the parsed table as a single tibble. A ZIP resource is unpacked and its first parseable file is returned by default; all_files = TRUE keeps every contained file in one of these formats as a named list — one element per file. The delimiter of CSV/TXT resources is auto-detected (comma, semicolon, tab, pipe, …), so both standard and European-style (semicolon/comma-decimal) files are handled without special configuration. Column types are inferred by vroom, but seeded by default from data.gouv’s own csv-detective profile (use_tabular_types = TRUE) and you can force specific columns with col_types = c(col = "Date") (e.g. to make a mostly-padded ISO date column with a few non-padded stragglers read as Date, turning the stragglers into NA; explicit col_types always win). The profile is a best-effort signal — it only exists for single-file resources indexed by the tabular service, and a missing profile (or a ZIP member) falls back to type inference; pass use_tabular_types = FALSE to disable seeding entirely. Any parsing issues are attached to the returned table as an rdatagouv_problems attribute — read them with dg_problems(tbl) — rather than a noisy per-cell warning.
The discovery catalog (dg_find_datasets()) is restricted to the official tabular formats data.gouv.fr itself indexes (csv, csv.gz, xls, xlsx, parquet), so every listed dataset is in principle openable as a table. Use the format argument to narrow the catalog to datasets with a resource in a specific format (the v2 search API matches several formats as a server-side union). The exact n_resources/formats/has_table/has_schema resource columns are populated by passing resources = TRUE. Direct pulls additionally accept tsv, txt and json resources.
Disclaimer
This package is not affiliated with or endorsed by data.gouv.fr. It is an independent project, and the authors are not responsible for the content of the datasets it indexes. This first version is a proof of concept and may contain bugs. Please report any issues on the GitHub repository.
This package is the result of a joint effort initiated during the French Finist’R bootcamp which was held in Roscoff in August 2026. It served as a complex task to benchmark R package development assisted by AI. The refinements that came up during the bootcamp were made with the assistance of deepseek-v4-flash as provided by Albert API, the French government’s AI platform. The author is grateful to the Finist’R organizers and the Albert API team for their support and guidance.