🇫🇷 France RGE Installers — Certified Renovation Firm Leads

Every RGE-certified renovation company in France, from the government's own register — with the email address.

RGE (Reconnu Garant de l'Environnement) is the state certification a French firm must hold before a homeowner can claim MaPrimeRénov' or CEE subsidies for the work. It is, in practice, the list of every heat-pump installer, insulation contractor, solar fitter, window fitter and energy auditor in France that is allowed to do subsidised work — published by ADEME under the Licence Ouverte.

This Actor turns it into a lead list: one row per company, not one row per certificate.

---

📊 The numbers, measured on 2026-09-16

Every figure below was counted against the live API on the day this Actor was built. Nothing here is an estimate.

Qualification rows in the register	160,281
Distinct companies (SIRET)	~60,012
Rows per company	2.67 (one real firm holds 38 qualifications at once)
Register last rebuilt	daily — dataUpdatedAt was the capture day itself
Qualifications still valid today	159,807 of 160,281 = 99.70%
Full sweep	161 pages, ~3 minutes

Field fill — counted, not estimated

Two independent measurements, because a sample and a census answer slightly different questions.

A. Census — the register's own counts over all 160,281 rows (?qs=email:* etc., exact):

Field	Rows	% of rows	Companies	% of companies
email	158,399	98.83%	~59,581	99.28%
phone	154,194	96.20%	~57,616	96.01%
website	107,818	67.27%	~39,760	66.25%
email or phone	159,404	99.45%	~59,724	99.52%
neither email, phone nor website	66	0.04%	~33	0.05%

B. Sample — 1,000 rows pulled in true random order (sort=_rand), collapsed to 992 companies, recounted by offline_validate.mjs on every run:

Field	Fill		Field	Fill
siret	100.0%		email	98.8%
companyName	100.0%		phone	96.4%
streetAddress	100.0%		website	64.7%
postalCode	100.0%		latitude / longitude	100.0%
city	100.0%		departmentCode	100.0%
qualifications	100.0%		certifiers	99.9%
firstQualifiedOn / qualifiedUntil	100.0%		email or phone	99.2%

"Filled" excludes "", null, "N/A", "-", "Non renseigné" and "Inconnu" — the register uses the last two as real values, and counting them would overstate organisme and domaine.

The emails are real, per-firm mailboxes. 975 of the 980 addresses in the random sample are distinct (99.5%), so this is not one certifier's inbox repeated 60,000 times. 51.5% of them are free-mail domains (gmail 22.8%, orange 13.6%, wanadoo, hotmail, free…) — the expected signature of French sole-trader tradesmen, not a data defect.

---

🧩 What one row is — read this before you compare row counts

The register publishes one row per QUALIFICATION, not per company. A firm certified for heat pumps, wood stoves and condensing boilers appears three times. Across the whole register that is 160,281 rows for ~60,012 companies.

• oneRowPerCompany: true (the default) — one row per SIRET, with every qualification collapsed into a qualifications array plus qualificationCount, workDomains, certifiers and the overall validity window. You are charged once per company.
• oneRowPerCompany: false — the raw register grain, one row per qualification. Each row is charged, so the same company costs you about 2.67 times.

An actor that ships the raw rows and calls them "160,000 French renovation companies" is selling you 60,000 companies counted 2.67 times. The deduplication here is a streaming group-by over a SIRET-sorted cursor, so it is exact — a firm whose rows straddle a page boundary still comes out once, with all of its qualifications.

---

📦 What you get per company

Field	What it is
siret / siren	The French company identifier. siren is null when the register filed a placeholder.
siretIsValid	Luhn checksum on the SIRET. All 1,000 random-sample rows pass; the failures are the handful of foreign contractors.
companyName	Trading name as filed
email / phone / website	Contact details, normalised (email lower-cased and regex-checked, website given a scheme)
streetAddress / postalCode / city / fullAddress	Postal address
departmentCode	Derived: 75, 69, 971… and 2A / 2B for Corsica
latitude / longitude	Geocoded by the register. Null when absent — never a fabricated 0,0.
qualificationCount	How many RGE qualifications this firm holds
workCategories	Travaux d'efficacité énergétique, Installations d'énergies renouvelables, Etudes énergétiques, Rénovation globale
workDomains	The specific trades — heat pumps, external wall insulation, solar PV, wood boilers, energy audits… (31 values)
certifiers	qualibat (95,848 rows), qualitenr (43,643), opqibi (8,665), cnoa (7,849), qualifelec (3,552)…
certificates	QUALIBAT-RGE, QualiPAC module Chauffage et ECS, Qualibois Eau, QualiPV 36…
worksForHomeowners	The register's particulier flag — takes private-homeowner (MaPrimeRénov') work. 96.45% of rows.
hasCurrentQualification	Any qualification still valid today
firstQualifiedOn / qualifiedUntil	The firm's overall RGE window
qualifications[]	Every certificate: name, code, trade, category, certifier, validity dates, isCurrent, and the PDF certificateUrl
registerUrl	Link back to this firm's own rows in the live register

---

🔎 How to filter

All filtering happens at the source — rows you filter out are never fetched, never delivered and never charged.

• postalPrefixes — ["75"] Paris, ["69"] Rhône, ["971"] Guadeloupe, ["75001"] one exact postcode. Corsica has no numeric code, so "2A" and "2B" are expanded for you (2A = 200xx/201xx = 405 rows, 2B = 202xx–206xx = 461; together exactly the 866 rows of 20*).
• workDomains / workCategories / certifiers / certificates — picked from the register's own vocabulary, read live from its facet endpoint.
• keyword — full-text across every field. pompe a chaleur matches 21,061 rows.
• centreLatitude / centreLongitude / radiusKm — radius search. 10 km of central Paris = 6,271 rows; 10 km of central Lyon = 2,061.
• onlyCurrentQualifications (on by default) — drop lapsed certificates.
• worksForHomeownersOnly — only firms that take private-homeowner work.
• requireContact — only firms with an email or a phone, filtered at the source so you never pay for a row you cannot act on.

A worked example, measured live on 2026-09-16: Rhône + heat-pump installers + homeowner work + currently valid + reachable → 408 qualification rows collapsing to 407 companies, 100% with an email, 99.3% with a phone, 91.4% with a website — one request, 0.43 s, $0.37.

---

💰 Pricing

$0.0009 per row delivered — $0.90 per 1,000 companies. No run-start fee.

Scope	Rows	Cost
One department's heat-pump installers (Rhône, example above)	407	$0.37
1,000 companies (the default cap)	1,000	$0.90
Every RGE company in France	~60,012	~$54

You are charged for rows delivered. Rows excluded by your filters are never delivered and never charged, a run that matches nothing costs nothing, and delivery and billing are the same call — so a buyer who hits their charge cap stops receiving rows at exactly the point they stop paying for them.

With oneRowPerCompany off you are charged per qualification row, which is ~2.67× more for the same firms. The default is on.

---

⚖️ Source and legality

• Source: data.ademe.fr — Liste des entreprises RGE, published by ADEME (the French Environment and Energy Management Agency), the register behind the official France Rénov' installer search.
• Licence: Licence Ouverte / Open Licence (Etalab) — explicitly permits reuse, including commercial reuse, with attribution.
• robots.txt: data.ademe.fr/robots.txt is User-agent: * / Allow: /.
• No credentials, no login, no API key. Nothing here is behind an authentication wall.
• This is business contact data published by companies for the purpose of being contacted about renovation work. It is not personal data about consumers. Under GDPR you still need a lawful basis for B2B prospecting and must honour opt-outs — that is your responsibility as the data controller.

---

🛠 How it works, and why it is built this way

Cursor paging, never offsets. The API hard-caps size + skip at 10,000 and answers past it with an HTTP 400 whose body is plain text, not JSON — "size + skip" cannot be more than 10000. An actor that JSON.parses every response dies with a SyntaxError there. This one follows the envelope's own next cursor, which is uncapped (12,000 rows walked across 12 pages, zero duplicate row ids), and never builds an offset URL at all.

The cursor is followed verbatim. The server rewrites every next onto the numeric dataset id, so reconstructing the URL from the slug silently restarts the walk at row 0.

total appears on the first page only. Every cursor page omits it. It is read once, used to floor-assert the walk, and its absence is never read as zero.

Filter values are matched as exact keywords, not as phrases. The register really does publish Isolation par l'intérieur des murs ou rampants de toitures ou plafonds with two spaces — that exact string matches 18,199 rows and the single-space version matches zero. A near-miss returns an empty result that looks exactly like a genuine "no such firms". So every value you supply is checked against the register's live facet list before the run starts, and a bad value fails the run with the real list, having charged you nothing.

A non-200 is a transport failure, not the end of the data. Each page is retried on a fresh proxy session; if it still fails the run stops and reports the exact unreachable URL. It never skips a page and reports success, because a silently missing page in a register looks identical to a register that is simply smaller than you thought.

Short pages are caught. The API responds gzipped and chunked with no Content-Length, so bytes cannot be checked. Instead: a page shorter than requested while still offering a next cursor is treated as a truncated body and fails the run. (A short page with no cursor is the normal last page — measured on a 44-row query that ended 10/10/10/10/4.)

---

⚠️ Honest limits

• This is a certification register, not a company database. There is no revenue, headcount, legal form or director name here. Join on siren if you need those.
• ~33 companies in the whole register publish neither an email, a phone nor a website. They are still delivered and charged (they have a SIRET, a name and an address). Set requireContact: true to exclude them at the source. The exact count for your run is in RUN_SUMMARY.
• A firm's own rows occasionally disagree with each other. On the captured pages the address differed across rows for 9 of 124 multi-row firms, the website for 8, the email for 2. The merge takes the most recently started qualification's values first and back-fills nulls from older rows. There is no way to know which one ADEME considers current.
• departmentCode is derived from the postal code. 20xxx covers both Corsican departments, so the 2A/2B split uses the standard 200/201 vs 202–206 rule. It is a postal split, not an INSEE lookup.
• Coordinates are ADEME's, and they are not always right. The register geocoded a Portuguese firm to a point near Paris. Where the register publishes no coordinate the field is null, never 0,0.
• A handful of foreign contractors sit at the top of an unfiltered run. SIRET order puts the register's placeholder identifiers (00000000…, postal code 00000) first. They are flagged siretIsValid: false and their postalCode, departmentCode and siren come back null rather than as fake values. Use a filter if you want a targeted slice rather than the first N.
• The row count moves. ADEME rebuilds the dataset daily; it was 161,428 in an older survey and 160,281 on 2026-09-16. Nothing in this Actor hardcodes it.
• Not tested through an Apify datacenter proxy. The source has no anti-bot, no user-agent gate, no rate limit (20 concurrent requests → 20× HTTP 200) and robots: Allow: /, so an IP block is unlikely — but it was measured from a normal residential IP, not from Apify's pool. The proxy input is optional and the Actor runs fine with it switched off.

---

❓ FAQ

What is RGE? Reconnu Garant de l'Environnement — the French state's certification for energy-renovation contractors. A homeowner can only claim MaPrimeRénov' or CEE subsidies if the work is done by an RGE-certified firm, so RGE status is effectively a licence to do subsidised renovation work.

Who buys this list? Heat-pump, insulation, boiler and solar manufacturers recruiting installer partners; MaPrimeRénov' lead networks and renovation marketplaces sourcing supply; tooling, materials and software vendors selling into the trade; and anyone building a "find an RGE installer near me" product.

Is this the same list as France Rénov'? Yes — France Rénov's public installer search is built on this ADEME register. This Actor reads the register directly, so there is no scraping of the consumer-facing site.

Why is website only ~65% filled? Because two-thirds of these firms have a website and a third do not. French renovation work is dominated by sole traders and small artisan businesses that operate on a mobile number and a Gmail address. That is the market, not a gap in the data — 99.5% of companies still have an email or a phone.

How do I get one department only? postalPrefixes: ["31"] for Haute-Garonne. Several at once are OR-ed: ["75", "92", "93", "94"].

How do I get only firms whose certificate has not expired? onlyCurrentQualifications is on by default. 99.70% of the register is current, so this mostly removes the 474 lapsed rows — which are exactly the firms you do not want to pitch as RGE partners.

Can I get every RGE company in France in one run? Yes. Set maxCompanies: 0. That is ~60,012 companies, 161 pages and about three minutes.

Does it handle a company that holds 38 certificates? Yes — that is a real firm in the register. It comes back as one row with qualificationCount: 38 and all 38 in the qualifications array, and is charged once.

What is in RUN_SUMMARY? How many rows the register said matched, how many were actually read, how many companies were delivered (= charged), the rows-per-company ratio actually observed for your filter, the count of companies with no email and no phone, and the exact Lucene query that was run.

Do I need a proxy? No. The source has no anti-bot and no rate limit. The proxy input is there if you want it.