swift-webrtc

main

Star
WebRTC data channels in pure Swift
1amageek/swift-webrtc

swift-webrtc

A pure Swift implementation of WebRTC data channels.

Built entirely from scratch — no C/C++ WebRTC library dependency. Implements the full protocol stack required for WebRTC Direct data channel communication:

UDP → STUN / ICE Lite → DTLS 1.2 → SCTP → Data Channels

Requirements

  • Swift 6.2+
  • macOS 15+ / iOS 18+ / tvOS 18+ / watchOS 11+ / visionOS 2+

Installation

dependencies: [
    .package(url: "https://github.com/1amageek/swift-webrtc.git", from: "0.0.1"),
]

Architecture

The library is split into independent modules:

Module Description RFC
STUNCore STUN message encoding/decoding, MESSAGE-INTEGRITY, FINGERPRINT RFC 5389
ICELite ICE Lite agent for server-side connectivity checks RFC 8445
SCTPCore SCTP association, chunk encoding/decoding, stream management RFC 4960
DataChannel Data channel lifecycle, DCEP (open/ack) messages RFC 8831
WebRTC Top-level API integrating all layers —

DTLS is provided by swift-tls.

Usage

Creating an endpoint

import WebRTC

let endpoint = try WebRTCEndpoint.create()
print(endpoint.localFingerprint.sdpFormat)

Client

let connection = try endpoint.connect(
    remoteFingerprint: remoteFingerprint,
    sendHandler: { data in
        // Send raw bytes over UDP
    }
)

connection.setRemoteICECredentials(ufrag: remoteUfrag, password: remotePassword)
try connection.start()

let channel = try connection.openDataChannel(label: "data")
try connection.send(payload, on: channel.id)

Server

let listener = try endpoint.listen()

for await connection in listener.connections {
    try connection.start()

    for await channel in connection.incomingChannels {
        print("Channel opened: \(channel.label)")
    }
}

Design

  • Transport-agnostic — Callers provide a SendHandler closure and feed incoming bytes via receive(_:). This allows integration with any UDP transport.
  • Sendable — All public types conform to Sendable. Thread safety is achieved using Mutex<T>.
  • Modular — Each protocol layer is a standalone library that can be used independently.

Benchmarks

Performance benchmarks are included under Tests/PerformanceTests/. Each module has a dedicated benchmark suite:

Suite Coverage
SCTPBenchmarks CRC-32C, packet encode/decode, TSN tracking, fragment assembly
STUNBenchmarks Message encode/decode, FINGERPRINT, MESSAGE-INTEGRITY
ICEBenchmarks STUN request processing, credential generation, peer validation
DataChannelBenchmarks DCEP encode/decode, channel open/lookup

Running benchmarks

# All benchmarks (debug)
swift test --filter PerformanceTests

# All benchmarks (release — recommended for accurate numbers)
swift test -c release --filter PerformanceTests

# Single suite
swift test -c release --filter SCTPBenchmarks

Release mode is strongly recommended. Debug builds include bounds checks and disable compiler optimizations, resulting in measurements that do not reflect production performance.

Key results (Apple Silicon, release build)

Operation Throughput
CRC-32C (1500 B) 2.7 GB/s
SCTP packet encode 845K ops/s
SCTP packet decode 1.5M ops/s
TSN gap block computation 577K ops/s
STUN FINGERPRINT compute 2.9M ops/s
Fragment assembly (multi-chunk) 526K ops/s

CRC-32C uses a slicing-by-8 lookup table algorithm. Checksum validation avoids packet-level copies by computing the CRC with the checksum field treated as zeros in-place.

License

MIT

Description

  • Swift Tools 6.2.0
View More Packages from this Author

Dependencies

Last updated: Sun Feb 01 2026 04:32:55 GMT-1000 (Hawaii-Aleutian Standard Time)