Website or desktop app
What the Plexora website handles well, where the desktop app or a node beside the data is the better place, and the measured numbers behind each.
The website (viewer.plexoraapp.com) and the desktop app are the same Plexora: the same tools, the same project files. The website runs everything inside your browser tab, so a few things are bounded by what one tab can hold or do. Within those bounds there is nothing to choose; beyond them the website says so, once per project, and offers the better place.
The numbers below are the website's benchmark records as of 2026-11-07 (a 12-core desktop with an NVMe disk and an RTX 3080, in Chrome). The website decides with the same numbers. To see them for your own computer, open viewer.plexoraapp.com/selfcheck and run the check: about a minute, and nothing is sent anywhere.
| Situation | Website | Desktop app | A node beside the data |
|---|---|---|---|
| A pyramidal image (OME-TIFF, SVS, OME-Zarr) | Opens as on the desktop; tested to 73 GB | Opens | Opens where the file lives |
| A flat (non-pyramidal) TIFF | Builds the pyramid in the browser: within about 2 minutes up to about 33 GB of pixels on a 12-core computer, less with fewer cores (a 9.6 GB image took 28 s, a 38 GB one about 2½ minutes); a copy kept in the browser needs room for about half its pixels (Firefox keeps about 10 GB in all) | Any size | Any size, without copying the file |
| A cell table | Up to about 40 million values (cells × columns) for gating and analysis, e.g. 1.26 M cells × 29 columns; less on a device with under 8 GB of memory | No browser limit | No browser limit |
| A job over a dataset (apply a gate to every image, a QC run) | One image at a time, with the tab open; applying a gate takes about 0.75 s an image of 100 k cells (0.61 s on the desktop), so the website suggests another place once a job would take over 10 minutes | Runs with no browser open | Runs beside the data |
| Magic select | With WebGPU: 0.04–0.07 s a click. Without WebGPU: about 1 s a click (0.95 s) | 0.38–0.41 s a click on the processor | On the node's own hardware |
| An image on a slow network share | Says so when the last 50 tiles take over 150 ms (95th percentile) | Reads the same share at the same speed | About 35 ms a tile (95th percentile) from local disk |
| An image in a cloud bucket (S3, Google Cloud Storage) | Opens in about 7 s | Opens in 21–25 s | Opens where the bucket is |
| Starting up | The home page shows at once. Opening an image waits until the rest has loaded: about 14 s at 50 Mbps on the first visit, about 6 s on later ones; the status at the top right says Starting meanwhile | Installed once | — |
Choosing a file or a folder
The website reads only what you give it, and reads it where it is: nothing is uploaded. In Chrome and Edge, Browse opens the same file window as the desktop app, listing the folders you have given the website:
- Choose folder… gives it a folder. A folder that is one dataset (an OME-Zarr, a Xenium or Visium run, a SpatialData store, a DICOM slide) is chosen at once; any other folder is kept under Folders and opened, so everything in it is a click away next time. Choose file… picks one file.
- Drop a file or a folder on the Browse window, the import dialog, a path box or the home page, whatever its size.
- One file of a DICOM slide brings the rest of the slide's folder with it; outside a folder you have given, the website asks for that folder first.
- After the browser restarts, Chrome asks again before the website may read a folder: open it and press Reconnect. To take a folder away again, right-click it under Folders and choose Remove from Folders.
Firefox and Safari cannot give a website a folder to keep. There, Browse asks File or Folder first, drops work as in Chrome, and a later visit asks for each file again.
What the website tells you
When something is beyond these bounds, a short notice appears in the corner with the measured reason and what to do about it: Desktop app opens the project in the desktop app (or offers it, if you do not have it), Connect a node pairs a Plexora node running beside your data, and Continue here carries on in the browser. A notice appears once per project in a tab; closing it or choosing Continue here keeps it away for 30 days. The desktop app shows no notices.
A node beside the data
The website connects to a saved server itself: Settings › Remote servers, then Connect (see Connect from the website). It signs in from the page and starts a node on the server for you.
A node is Plexora running on the computer that holds the files (a lab server, a cluster job) and serving them to your browser. Start one with plexora node serve --dynamic --pair and open the link it prints (Settings › Data nodes › Pair a Node shows the same command); see Install with pip for installing Plexora on a server.