Retinal intelligence, reimagined
See beyond
the surface.
Linux x86-64 release in preparation · Windows coming soon
01Overview
A clearer view of retinal analysis.
Diabetic retinopathy is one of the leading causes of preventable vision loss, and its early stages often cause no symptoms at all. Regular retinal screening helps changes surface early, yet timely specialist review remains hard to reach in many underserved and remote communities.
Ocuvia is a research prototype that explores how AI can support that screening workflow. It organizes images, model outputs and reports for professional review. It is not an autonomous clinical diagnostic device.
Retinal screening
A multi-stage pipeline screens each fundus photograph and estimates severity when the analysis continues.
Explainable insights
Lesion maps, vessel visualizations and attribution heatmaps show what the models responded to.
Integrated research workflow
Patients, examinations, results and reports stay together in one local workspace.
02How it works
From image to insight.
Three clear steps take a fundus photograph from import to a research result you can review, revisit and export.
- 01
Upload
Select a retinal fundus photograph.
- 02
Analyze
The local multi-stage AI pipeline performs screening and, when applicable, lesion segmentation, vessel analysis and severity estimation.
- 03
Review
Examine available visualizations, saved research results and an exportable report.
03Multi-stage pipeline
Intelligence at every stage.
Ocuvia analyzes each image as a cascade. Every stage has a focused task, and later stages run only when the earlier ones call for them.
- StageI
Screening gate
An initial screening model reviews each fundus photograph and decides whether it continues to detailed analysis.
- StageII
Lesion and vessel visualization
Segmentation models outline candidate lesions and trace the retinal vessel network as visual layers for review.
- StageIII
Severity grading ensemble
Where the cascade proceeds, a five-fold ensemble of grading models estimates diabetic retinopathy severity.
Every output is a research estimate intended for professional review. None of them is a diagnosis.
04Visual analysis
Make the invisible visible.
Each analysis produces visual layers you can inspect side by side, from the original photograph to the regions that shaped the model's output.
Masks and heatmaps are model-derived research visualizations. They are not independent confirmation of pathology.
05Explainability
Understand what influenced the model.
Grad-CAM, short for gradient-weighted class activation mapping, highlights the parts of an image that most influenced a model's prediction. Ocuvia saves this attribution as a heatmap you can lay over the original photograph.
That makes the model easier to question. You can check whether it attended to plausible retinal structures, or to something unrelated such as glare, artifacts or the edge of the image.
Shows where, not why
Attribution reveals which regions mattered most to the model. It does not explain the underlying condition.
Saved with the analysis
Overlays are kept alongside each result, so they can be revisited later.
Never proof of a diagnosis
Use attribution to question model behavior, always alongside qualified professional review.
06Reports and workspace
Every finding, thoughtfully organized.
Examinations, results and visual outputs stay linked to the right record, ready to review, revisit or export as a structured research report.
Patient-linked examinations
Each examination stays attached to its record.
Research screening reports
Structured, exportable summaries of each analysis.
Lesion and vessel visuals
Visual layers kept alongside the findings.
Analysis history
Revisit earlier results and saved overlays.
Local encrypted record storage
Records stay on your computer, in encrypted local storage.
07Desktop experience
Built for a focused desktop workflow.
Ocuvia runs as a native application on your own computer, keeping images, controls and results together in one focused workspace.
Local-first analysis
The analysis pipeline runs on your own computer, so retinal images do not need to leave it.
Offline-capable where supported
Where a build supports it, analysis runs without an internet connection. Each build's release notes say what it needs.
Native desktop interface
A native window with sidebar navigation and a dark-mode review workspace.
GPU acceleration
Uses compatible GPU acceleration where available. Hardware requirements vary by build.
08Download
Ocuvia for your desktop.
Choose your platform and get the latest available research build.
No public build yet
x86-64
Linux
The first verified Linux build is being prepared. Its version, file size and SHA-256 checksum will appear here the moment it is published.
- Version
- Not provided
- Release date
- Not provided
- Package format
- Not provided
- File size
- Not provided
- SHA-256 checksum
- Not provided
- Release notes
- Not provided
- Operating systems
- Not provided
- Installation
- Installation guide for Linux
- Download source
- Not provided
x64
Windows
A Windows installer is not available yet. It will be listed here only once a real, verified build exists.
- Version
- Not provided
- Release date
- Not provided
- Package format
- Not provided
- File size
- Not provided
- SHA-256 checksum
- Not provided
- Release notes
- Not provided
- Operating systems
- Not provided
- Installation
- Installation guide for Windows
- Download source
- Not provided
- Every listed file shows its SHA-256 checksum whenever one is published.
- Download buttons point straight at the published release file, never at a repository page.
- Keep your security software on. No step on this site asks you to switch it off.
09Installation
Get started in minutes.
A few careful steps take you from a verified download to your first analysis.
Linux
Release being prepared- 01
Download the supported package
Get the Linux build from the download section, and use only the file listed there.
- 02
Verify its checksum
Compute the file's SHA-256 checksum and compare it with the value shown for that exact file.
- 03
Install or run it
Follow the requirements of the package format you downloaded. The installation guide covers each format.
- 04
Launch Ocuvia
Start Ocuvia from your applications menu, or from the file you saved, depending on the format.
Windows
Coming soon- 01
Download the verified installer
Get the Windows installer from the download section once a real build is published.
- 02
Verify the file and its publisher
Compare its SHA-256 checksum, and if the installer is signed, confirm Windows shows the expected publisher.
- 03
Run the installer
Follow the installer's own setup steps through to the end.
- 04
Launch Ocuvia
Open Ocuvia once setup has finished.
Verify a download
Run the command for your system, replacing file-name with the name of the file you downloaded. The result must match the SHA-256 checksum shown for that file exactly. If it does not, delete the file and download it again.
Linux · Terminal
sha256sum file-nameWindows · PowerShell
Get-FileHash file-name -Algorithm SHA256Keep antivirus, SmartScreen and your other security protections switched on throughout.
Read the full installation guide10Transparency
Clear about what Ocuvia is, and what it is not.
Trust in a research tool starts with plain statements about its purpose, its limits and how it treats your data.
Research prototype
Ocuvia is a research prototype. It is not intended to replace professional medical evaluation or to provide an authorized clinical diagnosis.
Local-first architecture
Images and records are processed and stored on your own computer, in encrypted local storage.
Clearly labeled AI results
Model outputs are presented as AI-generated research results and kept distinct from professional findings.
Qualified professional review
Results are meant to be read by qualified healthcare professionals, who remain responsible for any clinical decision.
Release integrity
Downloads are listed with their SHA-256 checksum whenever one is published, so you can confirm a file arrived intact.
Open documentation
Where technical documentation is available, it is shared openly and linked from this site.
DocumentationSoon(link not available yet)
11Questions
Questions, answered.
Plain answers about what Ocuvia does, how to install it and how to read its results.
Ocuvia is a desktop research application for exploring AI-assisted retinal screening. It brings screening, severity estimation, lesion and vessel visualization, explainability views and structured reports together in one local workspace.
No. Ocuvia is a research prototype. It is not intended to replace professional medical evaluation or to provide an authorized clinical diagnosis, and its results should always be reviewed by a qualified healthcare professional.
Ocuvia targets Linux on x86-64 processors, with a Windows x64 build to follow. The download section always shows exactly which builds exist right now, with their version and file details.
Ocuvia is designed to run its analysis locally, on your own computer. You need a connection to download it. Anything a specific build needs beyond that is described in its release notes.
It can use compatible GPU acceleration to speed up analysis. Hardware and driver requirements depend on the build, so check its release notes before installing.
Download the build for your platform, verify its checksum, then follow the steps for that package in the installation guide. None of the steps asks you to switch off your security software.
Compute the file's SHA-256 checksum and compare it with the value shown in the download section. On Linux, run sha256sum file-name in a terminal. On Windows, run Get-FileHash file-name -Algorithm SHA256 in PowerShell.
If the two values differ, delete the file and download it again from the official source.
Yes. The sections above are reserved for real Ocuvia screenshots and outputs. Where one has not been published yet, you see a clearly marked placeholder rather than a mock-up.
Bug reports will go through Ocuvia's GitHub repository, which will be linked here once it is public. Include your operating system, the Ocuvia version and the steps that lead to the problem. Never attach identifiable patient images or data.
No. Research results come from experimental models and are meant for studying screening workflows. Clinical results come from qualified professionals using validated methods and examinations. Ocuvia's outputs should never be treated as a clinical result on their own.
Get started
A deeper perspective starts here.
Explore Ocuvia's retinal intelligence research workspace.