Self-hosted multiplayer VR — your own metaverse server. WebRTC SFU · Real-time pose sync · Spatial audio · Asset CDN · Unity + Unreal SDKs.

---

🥽 Why vrcollab?

Photon costs money. Mirror requires you to host. Unity Multiplayer keeps shutting down. Building your own VR multiplayer from scratch takes 3 months.

vrcollab is the open-source, self-hosted answer:

• 🌐 Up to 50 concurrent users per room (tested on a single 4-vCPU VPS)
• 🎙️ Spatial voice via WebRTC (LiveKit-compatible SFU)
• 🤸 Sub-60ms pose sync at 90Hz for hand + head tracking
• 📦 Built-in asset CDN with on-demand streaming
• 🏗️ Drop-in SDKs for Unity, Unreal Engine 5, and WebXR
• 🔒 End-to-end encryption optional
• 🛡️ Anti-cheat hooks built into the protocol

---

⚡ Quick Start

1. Run the server

git clone https://github.com/kasimmj/vrcollab
cd vrcollab
docker compose up -d

You now have:

• Signaling/room manager at :7880
• SFU media servers at :50000-50100/udp
• Asset CDN at :8080
• Admin dashboard at :7881

2. Drop the SDK into your project

Unity:

# Add via Unity Package Manager
git+https://github.com/kasimmj/vrcollab.git?path=/sdks/unity

using VRCollab;

var room = await VRCollabClient.JoinRoom("metaverse-hilla", playerName: "Kasim");
room.OnUserJoined += user => Debug.Log($"{user.Name} joined");
room.OnUserPose += pose => UpdateAvatar(pose);

Unreal Engine 5: Drop the VRCollab plugin into Plugins/, then:

#include "VRCollabSubsystem.h"

auto* VRC = GetGameInstance()->GetSubsystem<UVRCollabSubsystem>();
VRC->OnUserJoined.AddDynamic(this, &AMyGameMode::HandleUserJoined);
VRC->JoinRoom("metaverse-hilla", "Kasim");

WebXR:

import { VRCollab } from '@vrcollab/client';
const room = await VRCollab.join('metaverse-hilla', { name: 'Kasim' });

---

🏗️ Architecture

                ┌────────────────────────────────────────┐
                │           SIGNALING LAYER (Go)          │
                │  Room manager · auth · token issuance   │
                └─────────────┬──────────────────────────┘
                              │
                  ┌───────────┴───────────┐
                  │                       │
              ┌───▼──────┐         ┌──────▼────┐
              │  SFU #1  │ ... N   │  SFU #N   │
              │ (Pion)   │         │  (Pion)   │
              └────┬─────┘         └─────┬─────┘
                   │                     │
              ┌────▼─────────────────────▼─────┐
              │       VR Clients (Unity/UE)     │
              └─────────────────────────────────┘

         ┌──────────────┐   ┌──────────────┐
         │  Asset CDN   │   │   Postgres   │
         │  (S3/MinIO)  │   │  (rooms/users)│
         └──────────────┘   └──────────────┘

• Signaling: custom Go service over WebSocket — issues JWTs, manages room state
• SFU: Pion (Go) — relays audio + low-bandwidth pose streams between peers
• Asset CDN: MinIO + on-the-fly transcoding for textures and meshes
• DB: Postgres for room metadata, user accounts, and audit logs

---

🎙️ Spatial Audio

Audio is positioned in 3D space using HRTF (Head-Related Transfer Function) on the client. The server only routes streams — it doesn't process audio.

Configuration:

audio:
  spatial: true
  hrtf_dataset: "ircam-2024"      # or "mit-kemar"
  max_distance: 30.0              # meters before silence
  rolloff: "inverse"              # "linear" | "inverse" | "exponential"
  voice_codec: "opus"             # 16kHz mono, 32kbps

---

🤸 Pose Sync Protocol

vrcollab uses a custom binary protocol over WebRTC data channels for pose sync. Every 11ms (90Hz), each client sends:

PoseFrame {
  user_id:    u32
  timestamp:  u64 (microseconds since session start)
  head:       Pose (position + rotation)
  left_hand:  Pose
  right_hand: Pose
  body_ik:    [16 × Pose]  (optional, for full-body)
  voice_amp:  f32          (for lip sync)
}

Total: ~120 bytes per frame, ~88 Kbps per user. The server uses dead reckoning to predict missed packets and delta encoding to reduce bandwidth by ~60%.

---

🛡️ Anti-Cheat Hooks

vrcollab provides protocol-level hooks:

• Server-side teleport validation — refuse poses with sub-second jumps > 5m
• Voice amplitude clipping — prevent voice volume exploits
• Rate limiting per RPC, per user
• Replay protection via sequence numbers + timestamps

Game-specific anti-cheat (aimbot detection, animation cancels) is up to your SDK integration.

---

📊 Capacity & Performance

Benchmarked on a 4-vCPU 8GB VPS (single node):

Metric	Value
Max concurrent users in one room	50
Max simultaneous rooms	200
Total bandwidth at 50 users	~5 Mbps
End-to-end pose latency (LAN)	22ms
End-to-end pose latency (WAN US-EU)	95ms
Audio glass-to-glass latency	110ms

For larger deployments, vrcollab supports horizontal SFU scaling — add more sfu containers.

---

⚙️ Configuration

config.yaml:

server:
  signaling_port: 7880
  admin_port: 7881
  jwt_secret: "${JWT_SECRET}"

sfu:
  rtp_ports: 50000-50100
  ice_lite: false
  external_ip: "${PUBLIC_IP}"

rooms:
  default_capacity: 30
  max_capacity: 100
  idle_timeout: 5m

assets:
  cdn_url: "https://assets.your-domain.com"
  storage: "s3://your-bucket"
  cache_ttl: 24h

auth:
  provider: "jwt"           # or "oauth", "anonymous"
  oauth:
    providers: [google, github]

---

🌐 Use Cases

• Collaborative design — architects walking clients through buildings
• Remote education — instructors teaching in shared 3D spaces
• VR film production — directors blocking scenes with remote crew
• Multi-user training simulations — medical, industrial, military
• Social VR communities — your own VRChat/Rec Room alternative
• Real-time motion capture — sharing performances between studios

---

🚀 Roadmap

• Core signaling + SFU
• Unity SDK
• Unreal Engine 5 SDK
• WebXR client
• Avatar system (Ready Player Me integration)
• Full-body IK over the wire
• Recording + playback (.vrcap files)
• Federation (rooms across multiple servers)

---

📜 License

Apache-2.0. See LICENSE.

---

Star ⭐ to build your own metaverse.