SproutVideo

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the video hosting I actually need instead of SproutVideo.

The economics decide the design: **transcoding is CPU and delivery is bandwidth, and bandwidth is the bill.** Build the pipeline, rent the network.

STACK
- Node 20+ with Fastify
- SQLite through better-sqlite3, WAL mode
- ffmpeg on a worker for transcoding
- Object storage plus a cheap delivery network — Bunny is the usual answer
- hls.js in the browser, or the native player where HLS is supported
- Caddy in front

THE DATA MODEL
- videos: id, title, source_path, duration_ms, status, visibility, created_at
- renditions: id, video_id, height, bitrate, path, bytes, status
- playlists: id, video_id, path — the HLS manifest
- views: id, video_id, viewer_hash, started_at, seconds_watched, max_position, device, referrer
- heatmap: id, video_id, second, count — **built from positions, and it is the most useful thing this data produces**

THE PIPELINE
- Upload direct to storage, presigned and multipart, never through your server
- Probe with ffprobe, then transcode to a ladder: 360p, 540p, 720p, 1080p, only up to the source resolution. **Never upscale** — it costs CPU and bandwidth to make the picture worse
- HLS with segments of a few seconds and a master playlist
- A poster frame, and a sprite sheet for scrub previews, which is one ffmpeg command and makes the player feel expensive
- A queue with a concurrency cap and a priority: a short video must not wait behind a two-hour one

DELIVERY AND COST
- Everything served from the network, never from your origin. The origin is hit once per segment per region
- Long immutable cache headers on segments; the manifest short
- **Measure bytes per view and set a per-video and per-account cap.** A single popular video on an uncapped bandwidth bill is the story everybody in this category has

PRIVACY, WHICH IS WHY BUSINESSES BUY THIS
- Signed expiring URLs with a token tied to the viewer's session, and the signature covering the path
- Domain restriction checked at the network edge, not only in your player. **A player check is a suggestion; anybody can request the manifest directly**
- An option to watermark the viewer's identity into the picture for genuinely sensitive material

THE PLAYER
- hls.js where needed, native where possible. Keyboard controls, captions, playback speed, and remembered position
- Captions from whisper.cpp, free, and required rather than optional — **a business video without captions excludes people and is unwatchable on a phone in public**

ANALYTICS WORTH HAVING
- Play rate, completion, and a retention curve from recorded positions
- The drop-off point is the finding: it says which forty seconds of your video lose the audience
- Hashed viewer identity, retention with a sweeper, and nothing sent anywhere else

WHAT MATTERS MOST
Edge-enforced signing and a bandwidth cap. One makes the privacy real; the other stops success becoming an invoice.