CARPL.ai

Where CARPL.ai sits

CARPL.ai is not a medical imaging algorithm and it is not a PACS replacement. It is the layer in between: a vendor-neutral platform that lets a hospital or imaging network find, test, deploy and monitor radiology AI from many developers through a single technical integration and a single viewer. The problem it addresses is concrete. There are now more than 700 FDA-cleared radiology AI tools on the market, and every one of them historically arrived with its own integration, its own viewer, its own contract and its own accuracy claims. A large hospital imaging department cannot sensibly evaluate that catalogue one vendor at a time, and almost none can tell whether an algorithm that performed well in a developer's study will perform on their own patients and their own scanners. CARPL positions itself as the governance and orchestration layer that makes multi-vendor AI adoption tractable. In my view that positioning is honest about where the value is: not in any single clever model, but in plumbing, validation and oversight.

The company behind it is CARPL.AI Inc., with an operational base in Cupertino, California and further presences in India, Singapore and Delaware. It raised a $10 million Series A led by the International Finance Corporation, announced in July 2026, with Stellaris Venture Partners participating; third-party profiles put total funding at roughly $16 million and headcount at around 84 people. That makes CARPL a mid-stage specialist vendor, not a startup operating out of a garage and not a division of a healthcare giant. Both of those facts matter when you are signing a multi-year platform contract. The ecosystem picture is the other half of the story: CARPL has integration partnerships with Agfa HealthCare (whose RUBEE offering is powered by CARPL), Intelerad, DeepHealth/RadNet, PaxeraHealth, Ramsoft and Philips in Brazil, plus a foundation-model partnership with HOPPR for teams that want to fine-tune radiology models on their own data. A June 2025 announcement with Korean developer PurpleAI shows the acquisition pattern for the catalogue as well: individual FDA-cleared point solutions plugging into CARPL's distribution rather than building hospital sales teams of their own.

What CARPL.ai actually does

The product has four moving parts. First, the marketplace: CARPL lists 300+ third-party AI applications from 100+ AI vendors, spanning body parts, modalities and disease groups, plus open-source algorithms and the option to train and test models on your own institutional data. Second, the Universal AI Viewer, a browser-based interface where radiologists read images alongside AI output, and can edit, annotate, accept, reject or modify it. It supports a wide range of output formats — probability scores, region-of-interest coordinates, secondary captures, structured reports, DICOM PDF, GSPS presentation states and DICOM segmentation objects — which is more varied than most single-vendor viewers handle. Third, the validation tooling: what CARPL calls its DEV-D framework (Discover, Explore, Validate, Deploy), where institutions can run candidate algorithms against their own historical data before buying anything, with performance summaries, AUC metrics, bias and fairness reports and head-to-head comparisons between competing tools. Fourth, deployment and monitoring: one integration point to the hospital's PACS and RIS, on-premise or on any major cloud, with post-deployment performance monitoring, report auto-population, worklist triage, urgency flagging and automatic series identification and routing to the right algorithm.

On the regulatory side, CARPL holds FDA 510(k) clearance K232891, granted 27 March 2024, as a Class II Medical Image Management and Processing System under 21 CFR 892.2050, and it advertises CE marking along with clearances or certifications claimed across ANVISA, TGA and HSA markets. The platform displays each hosted algorithm's own 510(k) number with a link to the FDA decision, and CARPL states it verifies clearance status before integration.

My testing notes

A note on method first: CARPL is enterprise software sold through sales conversations, so I could not run it against live clinical studies. What I did examine closely was the 510(k) documentation, the marketplace's own published material across time, and the platform's product surface as described in its regulatory filings and partner announcements. Three concrete observations came out of that.

First, the marketplace count is a moving target and CARPL's own pages do not agree with each other. The FDA-clearance announcement from May 2024 describes 110+ applications from 50+ vendors. A June 2025 partnership announcement says 175+ applications from 75+ vendors. The current homepage says 300+ applications from 100+ vendors in one section and 275+ in another. Growth is plausible and probably real — the funding announcement in mid-2026 repeats the 300+/100+ figures — but when a vendor's homepage contradicts itself within a single scroll, I treat all catalogue counts as approximate. Buyers should ask for the current list in writing.

Second, the FDA clearance is routinely oversold in summaries of this product, and reading the actual decision letter corrects the impression. K232891 clears CARPL as image management and processing software. The indications for use cover storing, transmitting, processing and viewing DICOM images. The clearance contains no diagnostic performance claim and validates none of the algorithms running on the platform; each of those carries its own separate regulatory status, and the 510(k) summary is explicit that CARPL merely displays third-party outputs without modification. The letter also restricts what can be read on it: lossy-compressed mammograms and digitised film must not be used for primary interpretation, and mammography must be read on compliant displays. This is still a meaningful clearance — it is what allows FDA-cleared algorithms without their own cleared viewers to run inside a PACS workflow with human-in-the-loop review, which several clinical advisers cited in the announcement described as what actually opened the door to adoption. But 'FDA-cleared platform' and 'FDA-validated AI' are different statements, and CARPL's marketing walks close to the line between them.

Third, the workflow details in the regulatory filing are more informative than the marketing pages. The gateway-on-premises architecture, the study worklist with supervisor assignment to radiologists, the accept/reject/modify interaction on AI output, and the fact that customers technically cannot add algorithms themselves — only CARPL's compliance-vetted list — tell you this is governed clinical software with a controlled catalogue, not an open app store. That control is a feature for safety teams and a friction point for departments that want a specific niche algorithm tomorrow.

What CARPL.ai does not do

There is no public pricing, no self-serve trial and no free tier: every engagement goes through sales, which is normal for this category but means you cannot evaluate the commercial terms independently before committing staff time. The platform does not make any algorithm better, and it does not vouch for one; validation on your own data is the tool it gives you instead. It does not remove the per-algorithm regulatory homework — your team still needs to check each tool's clearance and intended use. It is not usable by an individual radiologist or a small practice that just wants to try one tool; the gateway, PACS integration and worklist design assume institutional IT. And despite the 'infrastructure agnostic' framing, an organisation deploying it is still choosing a long-term platform relationship with a company of roughly 84 people — a concentration risk worth discussing with procurement.

Verdict

Rated 3.9 on the ToolSpotter rubric — core functionality depth carries it (40% weight: the validation-before-purchase workflow and multi-format viewer are genuinely deep), value and usability are solid but unpriced and untestable from outside (20% each), the partner ecosystem is growing with named PACS integrations and marquee health systems (10%), and the regulatory and funding signals are verifiable and positive (10%). For a hospital imaging department drowning in per-vendor integrations, CARPL.ai is the most complete answer I have seen to the orchestration problem, provided you go in clear-eyed that the FDA clearance covers the plumbing, not the predictions.

Sources

• https://carpl.ai
• https://carpl.ai/news/carplai-raises-10mn-led-by-ifc-to-accelerate-healthcare-ai-adoption-globally
• https://carpl.ai/blog/carplai-receives-us-fda-clearance-to-make-clinical-ai-deployment-safer
• https://carpl.ai/news/carpl-partners-with-purple-ai-to-expand-access-to-fda-cleared-ai-for-intracranial-hemorrhage-detection
• https://fda.report/PMN/K232891
• https://fda.innolitics.com/device/K232891
• https://diagnosticimaging.com/view/fda-clears-radiology-ai-enterprise-imaging-platform-from-carpl-ai
• https://physicianaitools.com/tools/carpl-ai
• https://aitechsuite.com/tools/carpl.ai
• https://ai-market-watch.com/company/carpl-ai
• https://linkedin.com/company/carpl-ai
• https://insnaps.app/news/carpl-ai-audit