Browser capture, server-side compositing, two-tier storage
Recording is the easy part. Storage economics is the interesting one.
Try ZoomyStack
Browser-recorded WebM has no reliable duration. Storing every video twice is a bill that compounds.
The product is a self-hosted video platform for sales teams. A rep records in the browser, the server composites a branded deliverable with FFmpeg, and the published video is served to the prospect. Two constraints shaped the design. Browser-recorded WebM has no reliable duration — the container does not carry it, and the duration that JavaScript reports is often wrong. And storing every video twice (once as the raw recording, once as the published deliverable) is a bill that compounds with every rep and every call.
Browser capture
The rep records in the browser. The WebM is uploaded as a raw recording.
Composite
A serverless function runs FFmpeg to composite the branded deliverable from the raw recording.
Publish
The published video is moved to Cloudflare Stream. The origin copy is deleted.
The browser never talks to a model or a storage provider directly. The raw recording is uploaded to origin storage, a serverless function composites the branded deliverable with FFmpeg, and the published video is moved to Cloudflare Stream. The origin copy is deleted after a successful publish, so the long-term storage cost is the published video only. A design document that originally planned to keep the raw recording was reversed when production showed the cost.
The non-obvious calls — where the straightforward approach fails, and what it cost to find that out.
Why it breaks
Keeping the raw recording after publish doubles the storage bill for every video, forever. The raw recording is only useful for re-compositing, which is rare.
What I did
The origin copy is deleted after a successful publish to Cloudflare Stream. The design document that originally planned to keep it was reversed when production showed the cost.
What keeps it right once nobody is watching it run.
Browser-recorded WebM has no reliable duration. The composite measures the actual duration rather than trusting the value reported by JavaScript.
Whether you want this exact architecture deployed to your store, or a completely bespoke solution built from scratch.
Skip the R&D. I can deploy the Self-Hosted Video System architecture directly to your operations in days, not months.
Have a unique workflow? Let's build a bespoke app tailored perfectly to your complex business needs.