I'm Michael, a solo developer in Graz, Austria. I run SendVeil (https://sendveil.link/), a service for sending large original files through private, expiring links. This is a short write-up of the upload path, because "large files in a browser" turned out to be the hard part.
Parts, not one request
Every file is declared up front with its size and split into 64 MiB parts. Each part is uploaded as an R2 multipart part. The expected size of every part is computed on the server from the declared size, so a short or oversized part is rejected instead of silently corrupting the file. A 450 GB file is about 6,700 parts.
One Durable Object per workspace
State lives in a Durable Object per workspace, created with EU jurisdiction, backed by its SQLite storage. It records the multipart upload id once (a second caller gets the first id back, so a retry cannot start a parallel upload), the parts already accepted, and the file state (declared → uploading → complete). If the connection drops, the browser asks which parts exist and continues after the user selects the same file again.
Downloads
Recipients don't need an account. Downloads stream straight from R2 with HTTP Range support, so download managers and resumed downloads work, and the bytes are exactly what was uploaded: no compression or conversion.
What it is not
Transfers use HTTPS and R2 encrypts data at rest, but SendVeil is not end-to-end encrypted. I'd rather say that clearly than oversell it.
Try it
The Free plan gives 5 GB per transfer with 7-day retention and needs no card; paid plans go up to 450 GB per file. The UI is available in 20 languages.
Questions about the Workers/R2 side are very welcome.
Michael
Top comments (0)