We develop video calls in mobile apps. This is not just a WebRTC connection: camera and microphone management, system-level lifecycle, handling interruptions (incoming call, notification, screen lock), correct background and low-battery operation. Each of these details is a separate task with non-trivial solutions on iOS and Android. Our experience in this area exceeds 7 years, and we have delivered over 30 video call projects for clients from various industries.
Which Transport to Choose: WebRTC or a Ready SDK?
Pure WebRTC involves RTCPeerConnection, RTCSessionDescription, ICE/STUN/TURN servers, getUserMedia, signaling via WebSocket. A full implementation from scratch takes 3–6 weeks just for the transport layer, without the UI.
Most projects choose an SDK on top of WebRTC: Twilio Video, Daily.co, 100ms, Livekit, or Agora. They handle ICE negotiation, codec negotiation (VP8/VP9/H.264/AV1), adaptive bitrate, and provide ready native wrappers for iOS and Android.
If customization requirements are low and time-to-market is critical, we choose an SDK. If you need full control over codec, encryption, or minimal binary size, we opt for pure WebRTC via GoogleWebRTC (iOS) or org.webrtc (Android).
| Criterion | WebRTC | SDK (Twilio/Agora) |
|---|---|---|
| Development time | 3–6 weeks | 1–2 weeks |
| Control | Full | Limited |
| Binary size | Minimal | 5–10 MB larger |
| Codec support | Any | Limited by SDK |
How to Implement Video Calls on iOS?
Video from the camera is captured via AVCaptureSession. The correct initialization:
let session = AVCaptureSession() session.sessionPreset = .hd1280x720 let camera = AVCaptureDevice.default(.builtInWideAngleCamera, for: .video, position: .front) let input = try AVCaptureDeviceInput(device: camera!) let output = AVCaptureVideoDataOutput() output.setSampleBufferDelegate(self, queue: DispatchQueue(label: "videoQueue")) AVCaptureVideoDataOutput with a delegate provides raw CMSampleBuffer — we pass them to WebRTC via RTCVideoSource. Important: AVCaptureSession.startRunning() must be called strictly on a background thread, not the main thread — otherwise the UI freezes for 200–400 ms.
Switching between front and back camera: AVCaptureSession.beginConfiguration() + remove old input + add new + commitConfiguration(). Without begin/commit around these operations, video flickers during switching.
CallKit is mandatory for iOS apps with video calls. Without it, an incoming call appears as a push notification that can be missed. With CallKit, you get a full-screen system call screen, Bluetooth/AirPods integration, and correct audio interruption handling. We implement it via CXProvider and CXCallController. VoIP push via PKPushRegistry is the only way to wake the app for an incoming call.
How to Implement Video Calls on Android?
On Android, the Camera2 API (CameraManager.openCamera()) gives access to ImageReader with YUV_420_888 format — the standard format for WebRTC on Android. For most tasks, CameraDevice.StateCallback + CaptureRequest.Builder suffice.
ConnectionService is the Android equivalent of CallKit. We register a PhoneAccount and manage call state via TelecomManager. Without it on Android 10+, the app cannot stay in the foreground during a call without a persistent notification.
To keep the connection when the app is minimized, we use a ForegroundService with type mediaProjection or phoneCall. On Android 14, this requires explicit declaration in the manifest: android:foregroundServiceType="camera|microphone".
Camera and Microphone Management
Muting the microphone is not about disabling the device but replacing the audio track with silence. In WebRTC: RTCAudioTrack.isEnabled = false. Completely disabling the microphone at the AVAudioSession level breaks echo cancellation.
Flipping the camera during a call: call RTCCameraVideoCapturer.stopCapture(), change the device, and startCapture(with:fps:). Without a full stop, some Huawei and Xiaomi devices crash due to a race condition in Camera2.
Screen rotation: RTCVideoTrack itself does not account for orientation. You must pass RTCVideoRotation in the RTCVideoFrame during camera capture; otherwise, the remote party sees video rotated by 90° on devices with portrait-lock.
Connection Quality and Adaptation
Adaptive bitrate is a built-in feature of WebRTC (GCC algorithm). But you must explicitly set the range: RTCRtpEncodingParameters with minBitrateBps: 100_000 and maxBitrateBps: 1_500_000. Without an upper limit, on Wi-Fi WebRTC will try to use 4–8 Mbps — uncomfortable for other network users.
Connection quality indicator via RTCPeerConnection.getStats() — parse RTCInboundRtpStreamStats for framesPerSecond and packetsLost. Loss > 5% — show a warning to the user.
What Our Work Includes
- Requirements audit and optimal transport selection (WebRTC or SDK)
- Signaling development (WebSocket or third-party server)
- Camera and microphone integration with AVCaptureSession/Camera2
- CallKit/ConnectionService integration for system calls
- UI development: local video preview, remote video, control buttons
- Adaptive bitrate and connection quality monitoring
- Testing on real devices (iOS/Android)
- Integration documentation and source code handover
- Post-release support (2 weeks)
Process
- Requirements audit → 2. Transport selection → 3. Signaling implementation → 4. Camera/microphone integration → 5. CallKit/ConnectionService → 6. UI and control buttons → 7. Testing on real devices → 8. App store deployment.
We evaluate your project for free. Contact us, and we will propose the optimal solution considering your timeline and budget. A basic 1-on-1 video call using an SDK (Twilio/Agora/100ms) with camera control, mute, and flip takes 1–2 weeks. Pure WebRTC with signaling, CallKit, and background mode takes 3–5 weeks. The cost is determined after requirements analysis.