Live demo
Open source · MIT · Self-hosted

Your cameras, your server, your rules.

RTSPanda turns any RTSP/IP camera into browser live view, recordings and AI-powered alerts — one Go binary with an embedded React UI, a mediamtx relay and optional YOLO or Frigate detection. Runs on a server or a Raspberry Pi.

MIT License Release Last commit Stars Go React FastAPI ONNX Runtime Docker Raspberry Pi
Demo admin — pre-filled on the login page
Email demo@rtspanda.test · Password DemoAdmin!2026
Simulated cameras in your browser. No real video, nothing uploaded.
RTSPanda camera view with live object detection boxes on a driveway camera
16k+lines of Go, TS & Python
3deployment modes
1binary, embedded UI
11tagged releases

What it does

The pieces of an NVR, without the bloat — and with detection you can actually run on modest hardware.

📹

Browser live view

Low-latency HLS via mediamtx, multi-camera wall, snapshots and fullscreen — no plugins.

🎯

AI detection

Local YOLO (ONNX Runtime, CPU) on sampled frames, every 30 s by default, or Frigate events via webhook. Ignore zones and per-class confidence cut false alarms.

💬

Discord alerts

Snapshot or clip attached, cooldowns, mention suppression and per-camera rules.

⏺

Recording

Per-camera recording toggle, saved as fMP4 in 1-hour segments; browse, download and delete from the UI. Automatic retention is not implemented yet.

🥧

Runs on a Raspberry Pi

Pi mode does ingest and view locally and offloads detection to snapshot AI (Claude/OpenAI) or a remote YOLO worker.

📈

Observability

Health and readiness endpoints, Prometheus metrics, stream reset and ARM thermal monitoring.

Try detection and ignore zones in the demo →

Pick your deployment mode

One codebase, three runtime profiles. Toggle them in the demo’s Settings to see how behavior changes.

ModeHardwareDetectionBest for
standardx86 serverYOLO on sampled frames (local or remote CPU worker)Full stack
piRaspberry Pi / ARMSnapshot AI or remote YOLO workerEdge ingest/view with alerts
viewerDesktop or serverNoneLive view and optional recording

Architecture

A single Go backend serves the API and the embedded React app, supervises the relay and talks to an optional detector.

  IP cameras ──RTSP──►  mediamtx relay ──LL-HLS──►  Go backend (REST API + embedded React UI) ──►  Browser
                                                         │            │
                                                         │            └─► SQLite · snapshots · recordings
                                                         ▼
                                       Detector: YOLO worker (FastAPI + ONNX) · Frigate · Snapshot AI
                                                         │
                                                         ▼
                                           Discord alerts · Prometheus metrics

Run it yourself

Docker Compose is the quickest path for a standard server.

  1. Clone and start
    git clone https://github.com/248Tech/RTSPanda.git
    cd RTSPanda
    docker compose up --build -d
  2. Open the UI at http://<host>:8080 and add a camera with its RTSP URL.
  3. Verify
    curl -s http://127.0.0.1:8080/api/v1/health/ready

Full README · Android (Termux) guide · Changelog