aioxtm Vision · Ophthalmology AIDeployed in production · LaelVision SARL · DR Congo

Fundus in.
Findings out.
In about
two seconds.

AI-assisted retinal screening behind one HTTPS call. The fundus or OCT camera a clinic already owns, turned into structured findings — DR grade, glaucoma risk, broad screening, layer segmentation. EU-hosted, process-and-discard.

No sign-up. The samples are de-identified scans from our own clinics.

13
AI engines
3
clinical sites · DRC
~2 s
median inference
EU
hosted · Frankfurt
What it does

Thirteen engines. One API.

Separate purpose-built engines for breadth of findings and for segmentation, behind a single endpoint — picked automatically by image type.

01

Fundus analysis

Diabetic retinopathy · glaucoma · quality

ICDR DR grading (0–4), glaucoma risk, automated gradability screening. Confidence scores, attention maps, and a 44-condition broad screen ride on every response.

44-condition screenattention mapsvessel + A/V masks
02

OCT analysis

Layer segmentation · classification

Layer-level segmentation of the retina with disease classification — normal, AMD, DME and more — validated against our reference device format.

layer segmentationdisease classification
03

Automated PDF reports

Multi-image · branded · async

Patient report PDFs combining fundus and OCT findings with attention overlays. An async queue absorbs bursts — poll the job, or take a webhook.

background queuewebhook callback
Live demo

Try it on a real scan.

This calls the same production API our clinics use — nothing staged. The samples are real de-identified scans from our own clinics; or upload your own fundus photo or OCT. Images are analyzed in memory and never stored.

Fig. 02 — live inferenceLive · production API

Research demo — not a medical device. Don't upload images with patient information burned in.

Results appear here: DR grading, glaucoma risk, 44-condition screening, vessel maps and the model's attention overlay — in about 3 seconds.

How it works

From image to findings, in one call.

01

The clinic POSTs an image

One HTTPS call. Pseudonymous study ID in, JSON or PDF out. Drop-in for anything that speaks HTTP — MedFlow, a custom EHR, or a ten-line Python script.

02

The engines run on EU servers

Hetzner Frankfurt. Thirteen engines, picked automatically by image type. Process-and-discard: images are decoded, inferenced, and gone.

03

Findings return in about 2 s

DR grade, glaucoma risk, quality, attention map, vessel mask. Or enqueue a multi-image patient report and poll the job — the PDF lands in 8–15 s.

Model architectures, datasets and benchmarks are shared under NDA in diligence.

Case study · 01

LaelVision — three sites, one platform.

LaelVision operates ophthalmology clinics across three sites in the Democratic Republic of Congo. Their MedFlow clinic system embeds aioxtm Vision directly in the consultation workflow — screening happens where the patient record already lives.

No new hardware, no second screen. The cameras the clinics already owned now feed the screening engines, and the findings land in the chart the clinician already has open.

Built by the same team that runs the clinic OS underneath it — and maintained the same way.

OperatorLaelVision SARL
Sites3 · Matadi, Tombalbaye, Matrix — DR Congo
IntegrationEmbedded in MedFlow — the clinic's own system of record
AccessOne contract · per-site keys · full audit trail
CommercialsFirst commercial contract signed · first revenue collected in 2026
StatusDeployed in production
Compliance

Built for procurement.

The posture, stated plainly — the paperwork exists before the sales call does.

EU-hostedHetzner Frankfurt. No image data leaves the EU.
Process-and-discardImages are decoded, inferenced, and discarded. The audit log holds hashes and metadata — never PHI.
DPA + SCCsA bilateral Data Processing Agreement is signed per clinic. SCCs cover the US–EU corporate channel.
Decision supportAI-assisted screening in support of a qualified clinician — not a stand-alone diagnostic device. The disclaimer rides on every response.
Regulatory pathCE Class IIa (MDR) pathway mapped and costed — the notified-body route, not a shortcut.
Pricing

One flat number, quoted to your volume.

Flat monthly billing, unlimited inference, NET-14. Priced on clinical imaging volume, not site count — from a single clinic to a hospital network.

Single site to network
Quoted to
your volume.

flat monthly · billed monthly · NET-14

Request a demo
All thirteen engines — fundus, OCT, FFA, vessel segmentation
Per-site keys · audit trail · usage rollups
Async PDF reports with branded output
Idempotency cache · 24 h result reuse
DPA included · EU-only data path
Email support

Hospital networks running 6+ sites can take the inference stack on-prem, same API surface — quoted separately. Tell us your setup below and you will have a number back the same week.

FAQ

The short answers.

Is aioxtm Vision CE-marked?+

Not yet. The product is positioned as decision support — it returns scores and findings to a qualified clinician who makes the diagnostic call. The CE Class IIa (MDR) pathway is mapped and costed; the notified-body route is the deliberate next step rather than a shortcut. Every API response carries the disclaimer to keep that posture explicit.

How does GDPR work? Where do images live?+

Servers are in Hetzner Frankfurt — an EU-only data path. Process-and-discard: images are decoded, inferenced, and discarded; we never persist PHI. The audit log stores SHA-256 hashes and metadata only. A bilateral DPA is signed per clinic; SCCs cover the corporate parent's US–EU angle.

What about HIPAA and non-EU clinics?+

We can sign Business Associate Agreements for US clinics on request. The architecture — process-and-discard, encrypted in transit, EU-isolated by default — already satisfies most HIPAA Security Rule requirements; the BAA is the legal layer on top.

Can we self-host?+

Yes — hospital networks running 6+ sites can take a containerized version of the inference stack inside their own firewall, with the same API surface. Pricing is bespoke; talk to us.

How fast is integration?+

If your clinic system can make an HTTPS request, integration is one afternoon: HTTP client, Bearer token, two endpoints — POST /v1/analyze and POST /v1/report. There is a Laravel reference client, a Python snippet, and a Postman collection.

What happens if a payment fails?+

Stripe retries with smart dunning over about 14 days. If every retry fails, the keys for that customer return 402 Payment Required until the invoice settles — no surprise outages, no surprise bills.

Request a demo

See it on
your own
scans.

Tell us about your clinic and we run aioxtm Vision on a handful of your own anonymised scans — your images, your results, with a sandbox API key inside 24 hours. No commitment, nothing to install.

[email protected]Replies within 24 hours · no follow-up unless you ask