chaosskill/sirene-toulouse

SIRENE Toulouse — Registered Businesses & Establishments

Geolocated, filtered extract of the active business establishments registered in Toulouse (INSEE commune code 31555), sourced from INSEE's Sirene register — the official French database of all businesses and their establishments.

• 157,911 active establishments
• Point geometry (WGS84 / EPSG:4326 — [longitude, latitude])
• Format: GeoJSON FeatureCollection (CSV export also suggested — see Formats below)
• Geographic scope: Toulouse city limits only (INSEE commune 31555) — not the wider Toulouse Métropole (37 communes)
• Snapshot date: ~June 2026 (one-time extract of a live, continuously-updated register — see Freshness below)

Why this exists

INSEE's full national Sirene database is enormous (millions of establishments, 80+ administrative columns) and awkward to filter down to a single city. This extract does that filtering for you: only active establishments physically located in Toulouse, with the 14 fields most useful for local economic analysis, business-density mapping, and sector clustering.

What's inside

Each feature is one establishment (établissement) — a physical business location, not a legal company. A single company (unité légale) can have several establishments (e.g. a retail chain with 5 Toulouse shops = 5 features, one shared or absent legal name).

Active establishments	157,911
Employers (caractereemployeur = Oui)	21,736 (13.8%)
Distinct NAF sections (broad sector)	20 of 21 possible
Distinct NAF activity codes (activiteprincipaleetablissement)	700+
Records with no postal code determined ([ND])	18.1%
Records with no declared name (see Known limitations)	35.0%

Formats

Shipped as GeoJSON. If you need tabular/CSV for Excel, pandas, or Kaggle's built-in dataset preview, flatten properties + split geometry.coordinates into lon/lat columns — every field is already flat (no nested objects), so this is a one-line pandas.json_normalize() / geopandas.read_file() job.

Known limitations

• ~35% of establishments have no name field at all (denominationunitelegale AND enseigne1etablissement both empty). This is not a data quality issue — it's INSEE's privacy rule for the public Sirene diffusion: individual entrepreneurs and sole proprietors who have not opted into public name disclosure have their personal name withheld by law. Expect this gap especially among the 141,934 "non-employer" establishments (auto-entrepreneurs, freelancers, etc.).
• 18.1% of records have codepostaletablissement = "[ND]" (non déterminé) rather than a real postal code. Treat "[ND]" as a missing value, not a literal code.
• 11 records (0.007%) have coordinates well outside Toulouse's actual boundary (up to ~30km off) despite being tagged with Toulouse's commune code — a known artifact of address geocoding in the source register. If you're doing spatial density work, clip to the official Toulouse commune polygon rather than trusting the raw points at the edges.
• codecommuneetablissement and etatadministratifetablissement are constant ("31555" and "Actif" respectively) across every row — they're kept only so you can trace a row back to the national source dataset if needed; don't expect variation.
• categorieentreprise (PME/ETI/GE) is only populated for ~62% of establishments — INSEE only assigns this classification above a certain size/legal threshold; smaller/simpler structures are left blank rather than defaulted to "PME".

Freshness

This is a point-in-time snapshot, not a live feed. Sirene itself is updated continuously by INSEE (new registrations, closures, moves). This extract reflects the state of the register around the file's build date — treat it as "recent," not "real-time." Re-extraction would be needed for anything time-sensitive (e.g. tracking recent business closures).

License & attribution

Source data: INSEE Sirene (Système d'Identification du Répertoire des Entreprises et de leurs Établissements), published as open data under the Licence Ouverte / Open Licence 2.0 (Etalab). Reuse, including commercial reuse, is permitted with attribution to INSEE as the original data producer. This extract adds filtering and field selection on top of that public data — it is not a competing or independent data source.

(If republishing on Kaggle/Hugging Face: confirm and cite the exact origin dataset/URL you pulled base-sirene-v3 from, since that traceability matters for licensing clarity — not included here as it wasn't recorded at build time.)

Suggested uses

• Local business density / commercial vitality mapping by neighborhood
• Sector clustering and co-location analysis (which activities cluster together)
• Combine with IRIS or cadastral-section boundaries for choropleth maps of economic activity
• Employer vs. non-employer establishment ratio by area (self-employment intensity)
• Time-series comparison if paired with a future re-extraction

About La Ville Rose

This extract was built for lavillerose.com, a free interactive map of Toulouse that helps five kinds of people — families, students, companies, property investors, and seniors — find the right neighborhood for their specific situation. Instead of one generic map, each persona gets its own layer set: families see schools/noise/safety, investors see price trends and zoning risk, companies see foot traffic and business density, and so on. All the underlying data (this SIRENE extract included) comes from public open-data sources, cleaned and joined into something usable on a map.

• Map: lavillerose.com — pick a persona, explore ~15 data layers per profile for free, no signup.
• Guides & reports: guide.lavillerose.com — per-neighborhood guides, plus paid detailed/custom reports for anyone who wants a specific recommendation rather than exploring themselves.

This SIRENE extract is one dataset in a series of open Toulouse layers (transport, schools, PLU zoning, noise, green space, etc.) released alongside the map. A more advanced, cadastral-section-level dataset that merges this and other domains into per-neighborhood scores is available as a paid product on the same site, for anyone who needs the cross-domain join rather than the raw layer.