Versions and limitations
Versions
| Component | Version | Where it is set |
|---|---|---|
| webrtc-sdk Android | io.github.webrtc-sdk:android:150.7871.01 | build.gradle.kts |
| webrtc-sdk iOS, macOS | webrtc-sdk/Specs 150.7871.01 binary, checksum pinned, through a local package | Package.swift |
| libwebrtc Windows | libwebrtc.m150.7871.03 (webrtc-sdk/libwebrtc release), SHA-256 pinned | libwebrtc.lock.json |
| Windows App SDK | 2.5.1, .NET 10, CommunityToolkit.Mvvm 8.4.2, Vortice 3.8.3 | Directory.Packages.props |
| Windows effects | ONNX Runtime 1.24.4 with Windows ML, models converted with tf2onnx 1.16.1, opset 17 | convert.sh |
| MediaPipe Android | com.google.mediapipe:tasks-vision:1.0.0 | build.gradle.kts |
| MediaPipe Web | @mediapipe/tasks-vision, models from storage.googleapis.com | package.json |
| Jetpack Compose | BOM 2026.06.01, material3 1.5.0-alpha18 | build.gradle.kts |
| Android build | AGP 8.13.2, Gradle 9.5.1, Kotlin 2.3.0, compileSdk 36, minSdk 24 | android/ |
| iOS, macOS | Deployment target 26.0, Xcode 26 | WebRTCDemo.xcodeproj |
| sfu-server | Go 1.25, pion/webrtc/v4 v4.2.22, gorilla/websocket v1.5.3 | go.mod |
All native WebRTC builds come from the m150 branch of the webrtc-sdk fork.
Build commands
text
signaling-server: npm install && npm run dev (port 4000)
sfu-server: go run . (optional, group calls) (TCP + UDP 4001)
go test -race ./...
web: npm install && npm run dev (http://localhost:5173)
android: ./gradlew :app:assembleDebug
ios / macOS: open ios/WebRTCDemo.xcodeproj; scheme WebRTCDemo or WebRTCDemoMac
windows: ./native/RtcShim/scripts/build-shim.ps1
dotnet run --project src/WebRtcDemo.App -p:Platform=x64
docs: cd docs && npm install && npm run devLimitations
These are known and mostly on purpose. It is a demo.
- 1:1 by default. A default room holds two people. Group calls need the optional SFU, with 8 people per room by default.
- No TURN server. Calls can fail behind symmetric NATs or strict firewalls.
- No reconnect. A closed signaling socket ends the call.
- Signaling isn't secured. Plain HTTP and WebSocket, no authentication. Anyone with the room ID can join.
- The E2EE key goes through the server in plain form. A real app should use a key agreement (for example ECDH) or a passphrase shared out of band.
- E2EE is chosen in the lobby and can't be toggled during a call. Both peers must choose the same setting.
- Group calls: no simulcast and no downlink adaptation. Every subscriber gets each publisher's single stream. A slow subscriber can't ask for a lower layer.
- Group calls: one server, IPv4 UDP only. Clients must reach UDP 4001 directly. Rooms live in one process's memory.
- Group calls: chat goes through the server in plain text over the signaling WebSocket, unlike the 1:1 data channel.
- iOS always sends VP8. It is encoded in software, so it uses more CPU and battery than hardware H264. That is what keeps screen sharing alive in the background.
- macOS screen sharing needs Screen Recording permission, and the app must be relaunched after granting it.
- Windows composites effects on the CPU. The models can use the GPU or NPU, but blending, blur and color conversion run in C# on one thread, about 13 ms per 720p frame, so a slow PC sends effects at a lower frame rate.
- Windows plays video files only in formats Media Foundation can decode on that PC.