CaptureKit

On Apple platforms, capturing audio and video seems straightforward on the surface. In practice, as soon as you move past the trivial case — microphone + M4A file — API fragmentation makes everything complex. AVCaptureSession for the camera, AudioEngine for the mic, ScreenCaptureKit for macOS screen capture, ReplayKit on iOS, VideoToolbox for H.264 encoding, AudioToolbox for AAC… Each component has its own API surface, its own threading constraints, its own buffer formats.

While building the Atelier Socle streaming ecosystem — HLSKit, RTMPKit, SRTKit, IcecastKit — the most fundamental building block was missing: capturing the signal. Every transport needed encoded audio buffers, compressed video frames, consistent permission management, level monitoring. And every app had to reimplement the same capture logic, with the same pitfalls.

CaptureKit unifies all of it. A single CaptureSession orchestrates sources, encoders and outputs. A single StreamingPipeline connects capture to any transport via a StreamingTransport protocol. All built with Swift 6.2 strict concurrency — actors, AsyncStream, Sendable everywhere — no Combine, no external dependencies, with native support for macOS 14, iOS 17 and visionOS 1.

CaptureKit covers the entire capture path: from the hardware source to the encoded output. Audio and video sources, hardware encoders, file or streaming outputs, real-time metering, device discovery, permission management — all within a unified API.

Record microphone audio as AAC to an M4A file:

Via Swift Package Manager, add the dependency to your Package.swift:

Then add it to your target:

CaptureKit is organized around three foundational protocols that compose freely. Each source produces an AsyncStream of raw buffers, each encoder transforms those buffers into compressed data, and each output consumes the encoded data. CaptureSession orchestrates everything.

Nine audio sources cover every use case — from the built-in microphone to a programmable tone generator. Each source is configured with an AudioSourceConfiguration and produces an AsyncStream<AudioBuffer>. MicrophoneSource exposes automatic gain control, echo cancellation, noise suppression and voice isolation.

AudioSourceConfiguration provides ready-to-use presets:

CameraSource handles built-in and external cameras with full control — zoom, torch, focus mode, exposure, white balance, stabilization, depth data, and photo capture during recording.

Video configuration presets:

ScreenCaptureSource unifies three backends under a single API. On macOS, ScreenCaptureKit provides display, window, application or region capture. On iOS, ReplayKit captures the current app, while Broadcast Extension enables full system capture via an App Group.

CaptureKit exposes 12 codecs — 6 audio and 6 video — each implemented as an actor with configuration presets. Video encoders use VideoToolbox for hardware acceleration, audio encoders go through AudioToolbox.

FileOutput records to 7 container formats with automatic rotation by duration and/or size. Metadata (title, artist, album) is embedded in the file.

StreamingPipeline is the centerpiece for live. It orchestrates capture, encoding and delivery in a unified pipeline. Three operating modes — audio only, video only, or muxed audio+video. In muxed mode, the pipeline waits for the first video keyframe, extracts parameter sets (SPS/PPS for H.264, VPS/SPS/PPS for HEVC), sends them via sendConfiguration, then starts the interleaved audio+video stream with a shared monotonic clock.

The StreamingTransport protocol is deliberately minimal — four methods to implement:

Concrete implementations — RTMP, HLS, SRT, Icecast bridges — live in the consuming app, keeping CaptureKit transport-agnostic. The demo app includes working bridges for all four protocols in the Atelier Socle ecosystem.

An RTMP bridge connects CaptureKit to RTMPKit for broadcasting to Twitch, YouTube, Facebook and other RTMP platforms:

For audio streaming to Icecast/SHOUTcast, the IcecastKit bridge sends AAC or MP3 in audioOnly mode:

AudioMeter provides real-time audio levels with five analysis modes and waveform data for visualization. It accepts buffers from any source and exposes two AsyncStreams — one for levels, one for waveforms.

25+ presets cover common use cases, from streaming platforms to professional workflows. Each preset automatically configures codec, bitrate, resolution and frame rate.

DeviceDiscovery monitors audio and video device connections and disconnections in real time via an AsyncStream<DeviceChangeEvent>. Each device exposes its capabilities, connection type and supported formats.

PermissionManager unifies permission checking and requesting for 6 resource types. Status changes are exposed via an AsyncStream<PermissionChange>:

During capture, the session allows hot-swapping sources, adjusting bitrates, forcing keyframes and monitoring 17 event types — from state changes to dropped frames to transport quality.

On visionOS, SpatialCameraSource captures MV-HEVC (Multiview HEVC) to produce stereoscopic spatial video, viewable natively on Apple Vision Pro. MVHEVCEncoder handles encoding with appropriate layouts.

The library is organized into clearly separated modules, each with a well-defined responsibility:

The test suite covers the entire library with unit, integration, end-to-end and showcase tests. Showcase tests illustrate recommended usage patterns.

The DocC documentation covers 17 guides:

CaptureKit is the capture building block of the Atelier Socle streaming ecosystem. It integrates with all four transport libraries via the StreamingTransport protocol: