IdeaWell

Project ideas from Hacker News discussions.

We replaced H.264 streaming with JPEG screenshots (and it worked better)

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Corporate Network Restrictions

Enterprise firewalls block WebSockets/SSE, MITM proxies break protocols, forcing HTTP-only fallbacks.
"Corporate networks also love to MITM their own workstations... no WebSockets and no Server-Side Events either" - egorfine.
"Corporate IT needs to die" - ris.

2. Simplicity of JPEG/MJPEG Over H.264

JPEG polling provides reliable, low-latency streaming on bad networks vs. buffering H.264; prior art like jsmpeg praised.
"so did they reinvent mjpeg" - petcat.
"You can get close to 1080p60 per core if you use a library that leverages SIMD" - bob1029.

3. Stream Text Instead of Video

For code/terminal output, send text deltas or capture keystrokes rather than screenshots/video.
"why not send text instead of graphics" - Thinks.
"Capture the text somehow and send that... you only need to send each new character" - bambax.

🚀 Project Ideas

EnterpriseFallback Streamer

Summary

  • A JavaScript library for real-time screen sharing that auto-detects network constraints (e.g., corporate firewalls blocking WebSockets/UDP) and falls back from H.264/WebRTC to efficient HTTP-polled JPEG/WebP screenshots or HLS, ensuring low-latency viewing of AI agents or remote desktops.
  • Core value: Reliable streaming on any network, including strict enterprise setups, without user intervention—solves "The constraint that ruined everything: It has to work on enterprise networks" (egorfine).

Details

Key Value
Target Audience Developers building AI agent viewers, remote desktop tools, or live coding platforms targeting enterprises
Core Feature Network probing + seamless fallback (WebRTC → WS → SSE → HTTP polling), adaptive FPS/bitrate for static content like terminals
Tech Stack WebRTC, WebCodecs, MSE/HLS.js for browser; Node.js/FFmpeg backend for encoding; QUIC/HTTP/3 optional upgrade
Difficulty Medium
Monetization Revenue-ready: Freemium SaaS (free OSS lib, paid hosted TURN/proxies)

Notes

  • HN would love the "upgrade from polling to Websockets when the network allows" (Jakob) and corporate proxy workarounds like "streaming http using a standard content-type" (kevlened).
  • High utility for AI coding watchers; sparks debates on UDP vs. HTTP/3 in enterprises.

TextDelta Terminal Streamer

Summary

  • Open-source service streaming terminal sessions or code editor diffs as compact text deltas (insert/delete ranges) over HTTP/SSE, with browser reconstruction into a rich terminal UI—avoids video entirely for zero-latency on bad networks.
  • Core value: Lightning-fast, bandwidth-efficient alternative to video for "robot writing code" (robrain), perfect for mostly-static coding agents: "Capture the text somehow and send that... only need to send each new character" (bambax).

Details

Key Value
Target Audience AI agent platforms, remote SSH/VSCode users, terminal sharing tools like asciinema
Core Feature Real-time text diff streaming + cursor/scroll sync; browser renderer with search/copy; fallback to full dumps
Tech Stack xterm.js for client render; diff-match-patch lib; SSE/WebSocket over Node.js/Rust; integrates with tmux/screen
Difficulty Low
Monetization Hobby

Notes

  • Addresses "Why not send text instead of graphics?" (Thinks) and asciinema mentions (Snild); HN text-purists would rave.
  • Practical for global users (e.g., Huygens-like low-bandwidth); fosters discussions on semantic streaming vs. pixel-pushing.

CorporateVNC Adapter

Summary

  • Self-hosted proxy service converting VNC/RDP/neko streams to enterprise-safe HTTP-only formats (MJPEG/HLS over polling/SSE), with diff-based tiling for text-heavy screens and auto-bitrate adaptation.
  • Core value: "VNC-style approaches" (adamjs) made proxy-proof, solving "Corporate networks also love to MITM... no WebSockets" (egorfine) for low-motion screens like AI coders.

Details

Key Value
Target Audience IT admins/sysadmins deploying remote access in locked-down corps; AI platforms needing robust sharing
Core Feature VNC input → adaptive MJPEG/HLS output; tile diffs + CopyRect for static areas; MITM evasion via standard HTTP
Tech Stack noVNC/TigerVNC backend; FFmpeg/GStreamer for transcoding; Nginx proxy for HTTP/443 tunneling
Difficulty Medium
Monetization Revenue-ready: Open core + enterprise SaaS ($/user)

Notes

  • Quotes like "Copying how VNC does it is exactly how my first attempt would go" (klipklop) and bluescreen.live (karhuton) show demand; HN remote-desktop fans would upvote.
  • Utility for legacy systems + new AI tools; great for "kill corporate IT" rants turning productive.

Read Later