StreamHub Agent

Any camera in. AI on the stream.

StreamHub Agent is a lightweight edge agent — a tiny binary written in C — that runs on CCTV/IP cameras and small devices, pulls the RTSP feed and streams it to your StreamHub instance. Detection runs remotely on the server: YOLO objects, privacy masking, events pushed back to the device.

  • ARM
  • Linux
  • Windows
How it works

The edge streams, the server thinks

Follow one frame end to end: the agent pulls your cameras over RTSP, pushes a single RTMP/RTMPS stream across the internet to your StreamHub instance, the AI plugins detect remotely — and the results flow both ways: events back to the agent, live video out to every viewer.

StreamHub Agent topology: CCTV and IP cameras are pulled over RTSP by the StreamHub Agent, a tiny C binary running on an edge device (ARM, Linux or Windows). The agent pushes a single RTMP/RTMPS stream across the internet to a StreamHub instance, where AI plugins — yolo and deface — run detection remotely on the server. Detection events travel back over the same connection to the agent, while the StreamHub instance fans the live stream out to many viewers over WebRTC and HLS.

How StreamHub Agent connects cameras to a StreamHub instance Animated diagram: CCTV cameras feed the StreamHub Agent over RTSP, the agent streams RTMP/RTMPS across the internet to a StreamHub instance where AI plugins run detection, detection events return to the agent and the live stream fans out to viewers. INTERNET RTSP pull RTMP · RTMPS detection events WebRTC · HLS CCTV / IP cameras StreamHub Agent tiny C binary · ARM · Linux · Windows StreamHub instance SFU · API · recording AI PLUGINS — REMOTE DETECTION yolo deface Viewers
Capture

The agent pulls the RTSP feed straight from your existing cameras — no firmware changes, no re-cabling.

Stream

One RTMP/RTMPS uplink crosses the internet to your StreamHub instance — encrypted, firewall-friendly, no ports opened on site.

Detect remotely

YOLO and deface run on the server, not the device. Events flow back to the agent while every viewer watches live.

Platforms

One agent, three platforms

The same static binary everywhere: cross-compiled C, no runtime, no dependencies. If it can reach the camera and the internet, it can be a StreamHub edge.

ARM

Luckfox · Raspberry Pi · routers

Small enough for boards with a few MB of flash — run it right next to the camera, or on the camera itself.

  • ARMv7 and AArch64 builds
  • Runs on Luckfox-class boards
  • No GPU, no NPU required

Linux

x86_64 · servers · NVRs

Drop it on any box that already sits near your cameras: an NVR, a mini-PC, a spare VM on site.

  • Single static binary
  • systemd service out of the box
  • One agent, many cameras

Windows

desktops · site PCs

The PC that already runs the camera vendor software can also feed StreamHub — no extra hardware.

  • Native Windows build
  • Runs as a service
  • Same config, same one-liner
Remote detection

Why the AI lives on the server

No GPU on the edge

Cameras and Luckfox-class boards can't run YOLO. They don't have to: the agent only captures and streams — your StreamHub instance does the inference.

A binary that fits anywhere

Because there is no model on the device, the agent stays a tiny C binary: quick to install, cheap to run, happy on hardware from a decade ago.

Models upgrade centrally

Swap YOLO versions, tune privacy masking, add plugins — on the server, once. Every connected camera gets smarter without touching a single device.

Install

One line on the device. Live on your StreamHub.

Run the installer on the edge device, point it at your camera and your StreamHub instance, and the stream is live — detection included.

edge device — any platform
$ curl -fsSL https://streamhub.studio/agent.sh | sh
  • Detects your platform automatically
  • Tiny static binary, no dependencies
  • Reconnects and recovers on its own

Point a camera at it

Install a StreamHub node in one command, then let the agent bring your cameras in — or talk to us about managed hosting.