Introduction

ForgeVis is a high-performance RTSP video recorder and restreaming server. Record from IP cameras and restream to multiple clients using a single connection per camera.

Why ForgeVis?

• Efficient Architecture: Single connection per camera with broadcast distribution
• Fragmented MP4: Files playable while recording, streaming-optimized format
• RTSP Restreaming: Built-in RTSP server for distributing streams to clients
• Simple Configuration: YAML-based setup with minimal required settings
• No Transcoding by Default: video from the camera is recorded and served as is. Transcoding starts only when a viewer picks 720p over HLS

Key Features

Video Recording

• RTSP ingestion with automatic reconnect
• H.264 and H.265, AAC, G.711 and Opus audio
• Fragmented MP4 (fMP4) playable while recording, or plain MP4
• Configurable segment duration (default 900 seconds)
• Path templates with %path, %Y, %m, %d, %H, %M, %S placeholders
• Hooks on segment creation and completion
• Archive cleanup by retention period

RTSP Restreaming

• Full RTSP protocol server (OPTIONS, DESCRIBE, SETUP, PLAY, TEARDOWN)
• RTP over TCP (interleaved) and UDP
• H.264 (RFC 6184) and H.265 (RFC 7798) packetization
• Multiple clients per camera via broadcast
• URL format: rtsp://localhost:8554/{camera_id}

Browser Viewing

• HLS and Low-Latency HLS: http://localhost:8888/{camera_id}/index.m3u8
• Quality choice: source stream, substream and a 720p step on request (variants.m3u8)
• WebRTC (WHEP) with minimal latency
• MSE over WebSocket

Archive

• Playback of a recording from any moment: the server finds the right frame itself
• Export of a stretch and time-lapse, including from recordings spread across nodes

Substreams

• Main stream for full-quality recording
• Substream for video walls and slow connections
• URL format: rtsp://localhost:8554/{camera_id}/sub

Cluster

• Several nodes, with cameras moved automatically when a node fails
• Node roles: recording (recorder) and delivery (streamer)

Configuration

• YAML configuration file with shared defaults and per-camera overrides
• Cameras from the file, from a conf.d directory or from a database
• Changes applied on the fly, without a restart

Architecture

ForgeVis uses a hub-based architecture for efficient resource usage:

Camera (RTSP) → StreamHub → Recorder (fMP4 / MP4 files)
                         ├→ RTSP Server (restream to clients)
                         ├→ HLS (browsers)
                         └→ WebRTC (browsers)

How it works:

1. StreamHub: Maintains single RTSP connection per camera
2. Broadcast: Distributes frames to multiple consumers efficiently
3. Recorder: Subscribes to hub, writes segments to disk
4. RTSP, HLS, WebRTC: Subscribe to hub and serve the stream to clients

All components share the same video stream without additional camera connections.

Idle Timeout

ForgeVis automatically disconnects from a camera when there are no subscribers:

• Configurable idle timeout (default 10 seconds for the RTSP server)
• Camera resources saved while nobody needs the stream
• Automatic reconnect when clients appear again

Use Cases

• Multi-client viewing: One camera feed to multiple viewers
• Recording + monitoring: Record to disk while streaming to clients
• Storage optimization: Fragmented MP4 for efficient playback and streaming
• Edge processing: Single connection to camera, multiple local consumers
• Recording and delivery on separate servers: recording servers + servers that serve viewers

Getting Started

Ready to try ForgeVis? Head to the Quick Start guide.

Resources

• Configuration Reference - Complete config options
• Clustering & HA - Multi-node deployment and failover