Interacting with a remote phone only feels right when the screen reacts to your touch without a visible delay. That is a streaming problem, and the transport you pick decides whether the session feels live or sluggish.
Why not just screenshots
Polling screenshots (or MJPEG) is simple but wastes bandwidth and adds latency: every frame is a full image, and input round-trips are uncoordinated. Classic VNC improves on that but was designed for desktops on trusted networks, not mobile video over the public internet.
Why WebRTC
WebRTC was built for real-time media between browsers. ZekaForge uses it to stream the device screen as an encoded video track while relaying your input back over a low-latency data channel:
- hardware-friendly H.264/VP8/VP9 encoding instead of raw frames
- adaptive bitrate that degrades gracefully on a weak connection
- NAT traversal via STUN/TURN so sessions work behind firewalls
- DTLS/SRTP encryption on every track by default
The pipeline
- A device worker captures the screen and encodes it.
- A media server (SFU) forwards the track to your browser.
- Your pointer, keyboard and gesture events travel back over the data channel and are injected on the device via ADB.
Keeping latency low
Co-locating device workers and the media server in the same AWS region, preferring UDP, and tuning the encoder's keyframe interval keep the glass-to-glass delay small. When UDP is blocked, a TURN relay over TCP keeps the session working at a slightly higher latency rather than failing.
Demo mode
Until a device-worker fleet or a managed service is connected, the console runs a labeled demonstration so the streaming UI and input handling can be built and reviewed end-to-end.