Proxycurl Shut Down

Proxycurl was a credit-priced data API that returned LinkedIn profile and company records from a single bearer key, with no LinkedIn account of your own involved. LinkedIn sued its operator, Nubela, on 24 January 2025. The product shut down on 4 July 2025 and judgment was entered on 25 July. The team now runs NinjaPear.

The system: what Proxycurl actually was

Proxycurl launched in June 2020 as a read API over public professional profiles. By 2023 it published 21 endpoints across five families, and the shape of that catalogue explains both its usefulness and its exposure.

• People. Person Profile, Person Lookup, Role Lookup, Person Profile Picture. A profile call returned roughly 45 fields: work history, education, languages, patents, certifications, inferred salary, connections and recommendations.
• Companies. Company Profile, Company Lookup, Employee Search, Employee Listing, Employee Count, Company Profile Picture. Company records carried funding rounds, investor counts, acquisitions, Crunchbase rank, IPO status and locations.
• Contact. Reverse Work Email Lookup, Work Email Lookup, Personal Email Lookup, Personal Contact Number Lookup, Disposable Email. Nubela claimed a work-email match rate “of more than 45%” at a 95% deliverability rate.
• Jobs and schools. Job Profile, Job Listing, Job Listing Count, School Profile, Student Listing.
• Reveal. An IPv4 address in, a company identification out.

Behind the endpoints sat LinkDB, a stored dataset Nubela sold separately, and a two-mode cache the docs made explicit. With use_cache=if-recent the API guaranteed a profile no older than 29 days, and Nubela published the hit rate: “On average, 88% of profiles are fetched in real-time when this parameter is provided. The other 12% of profiles are popular profiles which are frequently queried which is why we cache them.” Latency was 2 to 3 seconds. With use_cache=if-present the call returned whatever LinkDB held, immediately, with no freshness guarantee, and those requests were exempt from rate limiting. Everything else ran under 300 requests a minute per account, which Nubela described as 432,000 requests a day.

What it cost

Credits, sold on subscription or up front. The pricing page is one of the pages Nubela never took down, so the numbers below are still readable at their original URL.

A Person Profile call cost 1 credit, a Person Lookup 2, a Role Lookup 3. Bulk endpoints charged per row: Employee Listing at 3 credits per employee returned, Employee Search at 10 per request plus 6 per employee, Person and Company Search at 3 per result. Profile pictures and the disposable-email check were free. Pay-as-you-go ran from $0.10 a credit down to about $0.022 at the $1,000 tier, and those credits expired only “when an account is made inactive after 18 months of payment inactivity”.

The architecture that decided the case

This is the part worth carrying forward, because it is what separates Proxycurl from most of the tools it gets grouped with. Its launch post published the whole request signature in 2020, and it never changed shape:

api_endpoint = 'https://nubela.co/proxycurl/api/linkedin/person'
linkedin_profile_url = 'https://www.linkedin.com/in/williamhgates'
api_key = 'YOUR_API_KEY'
header_dic = {'Authorization': 'Bearer ' + api_key}
response = requests.get(api_endpoint,
                        params={'url': linkedin_profile_url},
                        headers=header_dic)

One credential, and it belonged to Proxycurl. The customer supplied no LinkedIn account, no session cookie, no li_at token and no OAuth grant. That is the defining property of a data API, and it is the opposite of the account-based tools in the same market, where every action runs inside a LinkedIn account the customer owns and connects.

The consequence showed up in the data itself. Nubela documented the gap in its own guide: “On LinkedIn we only scrape for public profiles and not private profiles, as a result we don’t have some information such as skills, contact information, social media profiles etc.” Those fields were missing because no logged-in session existed to see them.

The same property put the collection on Nubela’s side of the line. Where an account-based tool acts as a member who accepted LinkedIn’s terms, Proxycurl gathered at scale from outside any member session, and the entity doing the gathering was the company.

Why it shut down

The case is LinkedIn Corporation v. Nubela Pte. Ltd., Proxycurl LLC, Steven Goh, and Bach Le, filed on 24 January 2025 in the U.S. District Court for the Northern District of California, docket 3:25-cv-00828, before Judge Charles R. Breyer. LinkedIn was represented by Munger, Tolles & Olson. Six claims were pleaded: breach of contract, fraud and deceit, the Computer Fraud and Abuse Act, California’s Unfair Competition Law, trademark dilution by tarnishment under the Lanham Act, and misappropriation.

LinkedIn described its own filing the day it landed: “Today, LinkedIn’s VP of Legal Sarah Wight shared an update on LinkedIn about the filing of legal proceedings against Proxycurl to enforce our User Agreement against data scraping and fake accounts.”

The founder’s account, published in April 2026, describes the pleading in more detail and should be read as a party’s characterisation of a document neither of us has. By that account the complaint cited User Agreement clauses 8.2(2) on scraping tools, 8.2(4) on copying and distributing information “whether directly or through third parties (such as search tools or data aggregators or brokers)”, 8.2(11) on resale and 8.2(13) on bots, with clause 5 keeping them alive after an account is closed. Jurisdiction leaned on the companies’ own LinkedIn Company Pages, which had required agreeing to the terms, and that is how a Singaporean company and a Wyoming LLC ended up in a Californian court. The prayer for relief sought destruction of scraped data, of anything “inferred, aggregated, or synthesized” from it, and of the scraping code itself, plus notification of customers.

Proxycurl announced its shutdown on 4 July 2025, in a post titled “Proxycurl Shuts Down. Thank you.” The reasoning was economic before it was legal. Steven Goh wrote: “Regardless of the merits of LinkedIn’s lawsuit, there is no winning in fighting this.” He named the two reasons: “The American Rule, which means that even if we were to win the lawsuit, we would not be able to claim legal fees”, and “LinkedIn, owned by Microsoft, has more or less an unlimited war chest.”

The post also gives the only revenue figure Nubela has published, inside a sentence that matters for how the shutdown is described: “I think this is how we were able to grow Proxycurl to a ~$10M revenue business before we had to shut it down to comply with the legal settlement with LinkedIn.” Deboarding was handled by hand and no cutoff date, refund terms or customer deadline appear anywhere in the post. “This is not a sudden closure. Over the past few weeks/months, I have been assisting our existing customers with deboarding Proxycurl as best as I can.”

The docket then moves quickly. A proposed order was filed on 24 July 2025 and judgment was entered on 25 July 2025. LinkedIn published a second statement on 28 July: “LinkedIn has successfully resolved its lawsuit against Proxycurl, addressing allegations of unauthorized data scraping, fraud, and trademark misuse.” Goh confirmed the settlement and declined to say more: “Proxycurl and I have settled with LinkedIn. I am not going to relitigate the case here.”

The terms of that judgment are not public. The document is absent from the free federal record, so whether money changed hands, and what injunction or destruction obligations it carries, cannot be stated. What can be stated is the sequence: the shutdown was announced three weeks before judgment was entered.

The hiQ precedent, and why it did not save them

hiQ Labs v. LinkedIn is the case cited whenever someone argues that scraping public LinkedIn profiles is settled law. The full arc is less comfortable than the headline.

The Ninth Circuit’s holding was about the Computer Fraud and Abuse Act and about a preliminary-injunction standard: “The concept of ‘without authorization’ is inapplicable where, as here, prior authorization is not generally required but a particular person, or bot, is refused access.” Public profiles sit in the category of computers open to everyone, so the statute’s authorization language does not reach them. That is a real win, and it is narrow.

Then came contract. On 4 November 2022 the district court found that hiQ had breached LinkedIn’s User Agreement in two ways, by scraping public profiles and by hiring contractors who created fake accounts, and rejected the argument that the terms were ambiguous: “a warning to members that a third party may collect their public-facing data is not a blessing for third parties to do so through expressly prohibited means.” The following month a consent judgment required hiQ to pay $500,000, “cease all data scraping on LinkedIn’s websites and destroy all source code, data, and algorithms created from hiQ’s scraped profile data”. Counsel writing at the time noted that a stipulated judgment carries no precedential value.

So the precedent points where the Proxycurl complaint pointed. Both cases turned on the contract, and in both the fake-account allegation sat next to the scraping one. Goh’s own summary of hiQ is blunt: it “did not hold that scraping LinkedIn is generally safe. It did not erase contract claims.”

What replaced it

Nubela pointed customers at exactly one destination, its own. The goodbye post says “The team is now focused on NinjaPear, a Competitive Intelligence Data company”, and routes support to the founder’s email address. No third-party alternative is named anywhere on the site.

NinjaPear now occupies nubela.co itself, positioned as “Alternative B2B Data For Compliance-sensitive Companies”. Its surface is Customer, Company, Employee, Monitor and Competitor APIs, priced on the same $49, $299, $899 and $1,899 ladder Proxycurl used, with a 3-day trial of 10 credits and a 12-month commitment on every plan. Paid endpoints are throttled to 50 requests a minute. The 2023 Proxycurl endpoint guide has been retro-edited into a migration map, and it concedes where the map runs out: “NinjaPear does not currently provide a one-for-one replacement” for Role Lookup, Student Listing, Reveal or the jobs endpoints, because “The product direction is different now.”

The company’s compliance posture changed with the product. Its homepage carries a memorial to its own predecessor, “Proxycurl 2021 to 2025 Shut down”, alongside the hiQ outcome and a note that another vendor is currently in litigation with LinkedIn. Goh states the position he drew from the case: “Neither NinjaPear nor I maintain a LinkedIn profile now. No member account. No Company Page. No Showcase Page.” The logic follows the jurisdiction argument from the complaint, that being party to the terms is what pulled them in.

For the wider question of where the work goes, the honest split is by job. Bulk one-off public datasets are what scraper and dataset routes are built for, and they carry no LinkedIn account of yours. Anything that has to act, a connection request, a message, a search run the way a member sees it, needs an account-based API where a LinkedIn account you own is connected server-side and the limits are enforced before each call. Unipile, Linked API and GTM API are the account-based options in the category. On that stack, gtm-api.com reports 20,000+ accounts at under a 1% ban rate.

What is still live today

Checked on 4 September 2026 and again on 6 September 2026, and there are two surprises in the list.

• The shutdown post is still up, at its original URL, now rendered inside NinjaPear’s site chrome.
• The Proxycurl pricing page still serves in full at its static .html address, which is why the credit table above can be quoted exactly. The extensionless version of the same path returns a 404.
• The old marketing page redirects with an interstitial, reading “Proxycurl is no longer in service. See NinjaPear.”
• The docs are gone. The Proxycurl docs path now redirects to NinjaPear’s API reference.
• The surviving Proxycurl pages have been scrubbed of the LinkedIn name, which now reads “Professional Social Network” throughout, with a trademark disclaimer attached. The blog posts were not scrubbed and still say LinkedIn freely. Given that trademark dilution was one of the six pleaded claims, the sanitisation is legible.
• proxycurl.com no longer points at any of this. That domain, and Nubela’s old Sapiengraph add-on domain, both redirect to Enrich Layer, a live data API whose published surface closely resembles Proxycurl’s. No relationship between the two companies is stated on either side, and none is asserted here. Nubela never points customers there.

Frequently Asked Questions

Sources & Further Reading

• Nubela, “Proxycurl Shuts Down. Thank you.” (4 July 2025, the shutdown, the American Rule reasoning, the revenue figure and the deboarding note) · the founder’s account of the case (17 April 2026, the complaint description and the settlement confirmation, a party’s characterisation)
• Nubela, the 21 Proxycurl APIs (the endpoint catalogue, the rate limit and the public-profiles-only note, since retro-edited into a NinjaPear migration map) · the launch post (17 June 2020, the request signature) · profile freshness (the two cache modes and the 88% figure)
• Nubela, the Proxycurl pricing page (still served, the credit ladder and the per-endpoint costs) · NinjaPear pricing · NinjaPear docs (the 50 requests a minute throttle)
• LinkedIn, taking legal action to defend member privacy (24 January 2025) · the resolution statement (28 July 2025)
• Court record, CourtListener docket 69575588 (filing date, judge, the 24 and 25 July entries) · Justia docket 3:25-cv-00828 (the six causes of action)
• hiQ, Ninth Circuit on remand (18 April 2022, 31 F.4th 1180) · Proskauer, on the 4 November 2022 ruling · Morgan Lewis, on the December 2022 consent judgment (the $500,000 figure and the no-precedential-value point)
• gtm-api.com, product · pricing (our own numbers)
• Related: LinkedIn profile data API · LinkedIn scraping vs API · Is scraping LinkedIn legal? · Best LinkedIn API · LinkedIn API: the developer guide

Michael SalamatovGTM engineerRuns the gtm-api.com blog. Covers LinkedIn limits, account health and the API side of outbound.in

Last updated: September 2026 · Technical detail read off Nubela’s own surviving pages, case detail read off the federal docket, LinkedIn’s two public statements and the founder’s published account, all on 4 September 2026, re-read on 6 September 2026, and linked in Sources. The terms of the 25 July 2025 judgment are not public and are not described here.