CalloraVoipSdk.Audio.Windows 4.6.0

CalloraVoipSdk

Commercial-grade .NET VoIP SDK for SIP signaling, RTP/SRTP media, WebRTC and PBX integrations.

CalloraVoipSdk is a .NET VoIP SDK (net8.0 / net9.0 / net10.0) for building softphones, PBX integrations, contact-center workflows and voice automation systems.
It exposes a stable, developer-friendly API through VoipClient while keeping transport, media and device internals behind a clean facade — and opens up through a module registry for building products like AI voice agents on top.

📖 Documentation: bechsteindigital.github.io/callora-voip-sdk 🧪 Examples: examples/ — runnable samples (BasicCalling, Dialer, Transfer, CustomAudio, VideoCalling, WebRtcPeer, WebRtcRecording, WebRtcDependencyInjection, and a browser video-call website WebRtcVideoCall.Web) 🛠️ Maintainers: MAINTAINING.md — architecture map, invariants, workflows; rules in ENGINEERING_RULES.md

Project status — 4.6.0. The SIP + RTP core is the mature, production-oriented surface: registration, in/outbound call control, transfer, DTMF, SRTP (SDES) and measured RTCP quality metrics — with symmetric RTP (comedia) as the production-proven NAT path. It is exercised in CI by an automated interop suite against a real Asterisk (PJSIP) container — registration, in/outbound calls with live RTP, codec negotiation (PCMU/PCMA/G722), SRTP-SDES, DTMF (RFC 4733), hold, blind & attended transfer, session timers (RFC 4028), early media (RFC 3960) and TCP/TLS transport, plus a two-leg bridged call with byte-exact bidirectional media verified through the PBX — the matrix runs with zero skipped cases. The WebRTC facade and DTLS-SRTP are validated in CI against real browsers (Chromium and Firefox, headless via Playwright: audio and VP8 video, SDK as offerer and as answerer), and the self-hostable STUN/TURN server is exercised end-to-end against a real coturn relay. Deliberate limits in this line: no data channels (SCTP), TURN relay is UDP-only, simulcast is send-side only, and full ICE (RFC 8445/7675) is opt-in and not yet production-proven — validate it for your trunk before enabling it. Known gaps and interop defects are tracked openly in the issue tracker — bug reports and interop feedback are especially welcome.

Contents: Why · Progressive API · Features · Install · Quickstart · Architecture · Contributing · Security · License

What's new in 4.6

• WebRTC facade (transport-only) — a signalling-neutral browser/peer surface in the CalloraVoipSdk.WebRtc namespace that mirrors the four-level design of VoipClient: WebRtcClient.CreatePeer() → IPeerConnection (ICE, DTLS-SRTP, BUNDLE, RTP/RTCP), a signalling happy path (peer.ConnectAsync(signalling, role)), the W3C track model (TrackReceived → RemoteTrack/EncodedFrame), a multi-peer manager (client.Peers), and L3 seams (IMediaTap, IWebRtcClientModule). The app owns signalling and the codec — the SDK moves bytes, it never encodes/decodes. Included: trickle ICE + early-bind (an ephemeral media port still yields a live m-line), mDNS .local candidate resolution (RFC 8828), NACK/PLI/FIR + RTX video repair (RFC 4585 / 5104 / 4588), transport-cc congestion feedback (RFC 8888), getStats, DTMF and RTCP quality on the BUNDLE path, and send-side simulcast (RFC 8853, offerer-confirmed). Validated in CI against real browsers — Chromium and Firefox, audio and VP8 video, SDK as offerer and as answerer. See the WebRtc* samples.
• Self-hostable STUN & TURN server — AddCalloraStunServer(...) / AddCalloraTurnServer(...) with TURN control over UDP/TCP/TLS, EVEN-PORT + RESERVATION-TOKEN (RFC 8656 §7), and a relay lifecycle that keeps allocations alive via permission-refresh and channel-rebind keepalives. Exercised end-to-end in CI against a real coturn.
• AEAD-AES-GCM SRTP/SRTCP (RFC 7714) — AEAD_AES_128_GCM and AEAD_AES_256_GCM end to end, offered preferred in DTLS-SRTP use_srtp with AES_CM_128_HMAC_SHA1_80 as the interoperable fallback.
• Early media (RFC 3960) — receive-only media from a 180/183 SDP, a pre-answer call handle (IPhoneLine.OutboundCallRinging), ICall.EarlyMediaSdp, and DTMF while ringing (IVR / AI outbound).
• SIP MESSAGE (RFC 3428) and SIP PUBLISH (RFC 3903) with the full soft-state lifecycle, plus REFER transfer progress subscriptions (RFC 3515 / 6665) and ICall.TerminationReason — a protocol-neutral end cause that tells busy, unanswered, cancelled and rejected apart from a generic failure.
• Local ICE restart — ICall.RestartIceAsync() (RFC 8445 §9): new credentials on the existing socket, role preserved, so an app can both detect a dead media path and repair it.
• Hardening across the stack — a full source audit closed every interop- and stability-critical finding; CI grew chaos/fault-injection and performance gates alongside the existing interop, soak and browser suites. See CHANGELOG.md for the complete list.

Known limits in this line: no data channels (SCTP), TURN relay is UDP-only (no TCP/TLS relay), simulcast is send-side only (receive-side RID demux is a later slice), Safari/WebKit is not yet verified, and ICE-TCP candidates (RFC 6544) are a deliberate omission — real trunk calls run over symmetric RTP (comedia), which needs no ICE or STUN.

Breaking change in 4.6 — the SIP-facade configuration types were renamed so each facade owns a facade-scoped name (parallel to WebRtcConfiguration/WebRtcOptions/AddCalloraWebRtc): SdkConfiguration → VoipConfiguration, SdkOptions → VoipOptions, AddCallora(...) → AddCalloraVoip(...). There are no compatibility aliases — rename these three symbols at your call sites. VoipClient and all other public types are unchanged; behaviour is identical. See CHANGELOG.md.

Why CalloraVoipSdk

CalloraVoipSdk is built for developers who need more than a black-box telephony wrapper.

• Stable public API centered around VoipClient
• Full SIP call control for outbound and inbound scenarios
• RTP media pipeline with sender, receiver and cross-connect support
• Runtime audio device control for Linux and Windows
• DDD-oriented architecture with clear boundaries
• Extensive RFC-oriented unit and compliance tests

Progressive API: simple first, deeper when needed

CalloraVoipSdk does not force an early choice between a black-box wrapper and a raw protocol stack. Start with managed workflows, then use deeper public contracts only where the product needs them:

The levels can be mixed per call; using a convenience method does not close off the lower-level contracts. SIP/RTP implementation classes and arbitrary wire mutation remain internal, so this is controlled extensibility rather than an unstable exposure of the entire transport stack. See Progressive API for the decision guide and the media-threading constraints.

Typical use cases

• Softphones and operator desktops
• PBX and SIP integrations
• Contact-center and queue workflows
• Voice bots and automation systems
• Media bridging and custom audio routing
• Real-time call control in backend services

Current feature set

Available in the repository today:

• SIP basics: register, invite/dial, accept, hangup, hold/unhold
• Advanced call control: DTMF, blind & attended transfer with REFER progress tracking (RFC 3515 / 6665, via TransferRequestedEventArgs.Subscription)
• Early media (RFC 3960): pre-answer receive-only media from a 180/183 SDP, a pre-answer call handle (IPhoneLine.OutboundCallRinging), the early SDP on ICall.EarlyMediaSdp, and DTMF in the early dialog (SendDtmfAsync while ringing — IVR / AI outbound)
• In-dialog operations: INFO, OPTIONS, SUBSCRIBE, NOTIFY
• Messaging & presence: SIP MESSAGE (RFC 3428) send & receive (VoipClient.SendMessageAsync / IncomingMessage event) and SIP PUBLISH (RFC 3903, e.g. presence) with the full soft-state lifecycle — PublishAsync plus RefreshPublicationAsync / ModifyPublicationAsync / RemovePublicationAsync (SIP-If-Match)
• Media stack: RTP sessions, sender, receiver, MediaConnector, cross-connect
• Media encryption: SRTP via SDES (RFC 4568) or DTLS-SRTP (RFC 5763, opt-in via VoipConfiguration.OfferDtlsSrtp) as both caller and callee, encrypted/authenticated RTCP via SRTCP (RFC 3711 §3.4), a configurable per-call policy (VoipConfiguration.SrtpPolicy: Disabled / Optional / Required), re-keying on re-INVITE, and the negotiated suite name / SRTCP status readable on ICall.MediaParameters
• Transport selection: choose the default outbound SIP transport (VoipConfiguration.DefaultTransport: UDP / TCP / TLS / WS / WSS; default UDP)
• Custom outbound headers (DialOptions.CustomHeaders, injection-guarded) and read-only remote identity on inbound calls (ICall.RemoteAssertedIdentity / ICall.Diversion)
• Per-call observability: ICE state + selected candidate pair (ICall.IceSnapshot) and raw RFC 3550 RTP counters (ICall.RtpStatistics)
• NAT/trunk controls: public signaling contact (SipAccount.PublicSipHost) and an opt-in public media address for CGNAT / static 1:1 NAT (SipAccount.PublicMediaHost)
• Self-hostable STUN and TURN server (RFC 5389 / RFC 5766 / RFC 8656) via AddCalloraStunServer(...) / AddCalloraTurnServer(...) — TURN control over UDP/TCP/TLS, EVEN-PORT + RESERVATION-TOKEN, permission/channel keepalives; the client side is verified in CI against a real coturn relay (UDP relay only)
• Per-call media tap: attach frame receivers/senders to any call for bots, bridging and streaming scenarios (client.Media.CreateReceiver()/CreateSender())
• Encoded video (transport-only): send/receive encoded frames (client.Media.CreateVideoReceiver()/CreateVideoSender()), a ready-to-use recommended outbound bitrate + NetworkQuality from transport-cc feedback, inbound keyframe flags and RTCP PLI/FIR keyframe-request feedback, plus a default-video convenience (client.AttachDefaultVideoAsync(call) with an application-supplied IVideoDevice codec). The SDK never encodes/decodes — bring your own VP8/H.264 codec
• Module registry (client.Modules) as the extension point for separately shipped feature modules
• WebRTC facade (CalloraVoipSdk.WebRtc): signalling-neutral browser/peer connections (WebRtcClient.CreatePeer()), an SDK-driven handshake (peer.ConnectAsync(signalling, role)), the W3C track model (TrackReceived/RemoteTrack/EncodedFrame), a multi-peer manager (client.Peers) and L3 media taps/modules — transport-only, bring your own codec (see What's new in 4.6)
• Configurable audio codec preference (VoipConfiguration.PreferredAudioCodecs)
• RTCP quality metrics with measured values: local/remote jitter, packet loss and round-trip time from SR/RR (LSR/DLSR); RFC 3611 XR-tolerant compound decoding
• Linux audio devices via CalloraVoipSdk.Audio.Linux
• Windows audio devices via CalloraVoipSdk.Audio.Windows
• Device codec support: PCMU, PCMA, G.722 and native Opus (RFC 7587, 48 kHz)
• Runtime device controls:
  • device hot-switch
  • input/output mute
  • input/output volume
  • format updates
• RFC-oriented unit and compliance test coverage

Package layout

• CalloraVoipSdk
  Public entry point and developer-facing facade

• CalloraVoipSdk.Core
  Core call, line, media and protocol abstractions

• CalloraVoipSdk.Audio.Windows
  Windows audio integration based on NAudio

• CalloraVoipSdk.Audio.Linux
  Linux audio integration based on PortAudio

Architecture

The solution follows a DDD-oriented structure:

• src/Core/Domain
  Entities, value objects, states and domain events

• src/Core/Application
  Use cases and orchestration for calls, lines and media

• src/Core/Infrastructure
  SIP, RTP, SDP and audio-specific implementations

• src/Client
  Public facade, convenience APIs and dependency injection wiring

Public API boundary

For SDK consumers, VoipClient is the central entry point, not the only useful abstraction. Its typed sub-facades and public domain/media contracts form the supported progressive API:

• VoipClient provides lifecycle ownership and managed workflows
• IPhoneLine and ICall provide typed call control and observable state
• public media, device, telemetry and module contracts are supported extension seams
• transport, parser and wire-level implementation classes remain internal details

This boundary keeps common integrations compact while still supporting advanced product logic without depending on internal implementation types.

Versioning

CalloraVoipSdk follows Semantic Versioning (MAJOR.MINOR.PATCH).

• Current public release line: 4.x — latest release 4.6.0 (see releases)
• Public API removals only happen in MAJOR releases; deprecations are introduced through [Obsolete(...)] before removal
• Consumer-relevant changes are documented in CHANGELOG.md

Requirements

• .NET SDK 8.0+ (packages target net8.0, net9.0 and net10.0)
• SIP account or PBX credentials
• For real audio I/O on Linux: CalloraVoipSdk.Audio.Linux
• For real audio I/O on Windows: CalloraVoipSdk.Audio.Windows

Installation

NuGet

dotnet add package CalloraVoipSdk
dotnet add package CalloraVoipSdk.Audio.Windows   # Windows
dotnet add package CalloraVoipSdk.Audio.Linux     # Linux

Local development via ProjectReference

<ItemGroup>
  <ProjectReference Include="..\voip\src\Client\CalloraVoipSdk.Client.csproj" />
  <ProjectReference Include="..\voip\src\Core\CalloraVoipSdk.Core.csproj" />
  <ProjectReference Include="..\voip\src\Audio\Linux\CalloraVoipSdk.Audio.Linux.csproj" />
  <ProjectReference Include="..\voip\src\Audio\Windows\CalloraVoipSdk.Audio.Windows.csproj" />
</ItemGroup>

Build and test

dotnet restore CalloraVoipSdk.sln
dotnet build CalloraVoipSdk.sln -c Release

# Architecture gates (CI runs these first)
dotnet test tests/CalloraVoipSdk.ArchitectureTests -c Release

# Standard test set (matches CI: excludes long soaks and Docker interop)
dotnet test CalloraVoipSdk.sln -c Release \
  --filter "Category!=SoakLong&Category!=Interop"

# Interop suite — real Asterisk (PJSIP) container via Testcontainers; needs a Docker daemon
# (tests self-skip when none is reachable)
dotnet test tests/CalloraVoipSdk.InteropTests -c Release --filter "Category=Interop"

# Linux workstation mode: runs Asterisk in Docker's host network without creating veth links.
# This avoids Chromium/Brave treating every test container as a host network change.
./scripts/test-interop-browser-safe.sh

# Long soak tests — media-quality drift + resource-leak/plateau guards over extended runs
dotnet test tests/CalloraVoipSdk.SoakTests -c Release --filter "Category=SoakLong"

# Manual machine-capacity envelope — real Asterisk echo, per-call/per-direction quality gate.
# Defaults ramp from 64 to 4096 calls and are intentionally not part of regular CI.
dotnet test tests/CalloraVoipSdk.InteropTests -c Release -f net10.0 \
  --filter "Category=Capacity"

Machine-bound capacity evidence (2026-07-28): on an 8-core/16-thread Intel i7-11800H host using Server GC, a restricted Office power profile and a co-located Asterisk Echo() container, 1,408 full-duplex calls passed the strict per-call/per-direction gate in independent runs. A 1,792-call stage passed once but crossed the p99 timing gate in a repetition; at 1,920 calls every call remained connected with complete RTP/RTCP evidence and at least 99% media delivery, while 27 calls recorded a 41-ms inbound p99 against the 40-ms strict limit. The first thing to fail was scheduler timing — not RAM or CPU. These numbers describe that host/profile, not a global SDK maximum; measure your own envelope with the same gate. Sizing guidance: Capacity and sizing.

The browser-safe interop runner is an explicit local Linux mode. It serializes Asterisk instances because host networking uses the fixed SIP ports 5060/5061 and RTP port range, disables the Testcontainers resource reaper only for that run, and removes only containers carrying its dedicated cleanup label. The normal command and CI continue to use the isolated Docker bridge.

Interop coverage. The interop suite runs the full SIP/RTP flow against a real Asterisk container in CI (currently all cases green, none skipped): registration (happy + failure), in/outbound calls with live RTP, codec negotiation (PCMU/PCMA/G722), SRTP-SDES media, DTMF (RFC 4733), hold/unhold, blind & attended transfer (REFER), session-timer negotiation (RFC 4028), early media (RFC 3960, pre-answer receive + DTMF in the early dialog) and TCP/TLS transport. A separate two-leg suite bridges two SDK legs through the PBX and verifies bidirectional, byte-exact media (RTP counters both ways, local + remote RTCP quality, and byte-identical PCMU payload A→B). The two-leg scenario matrix additionally runs against a real FreeSWITCH container through the shared IPbxFixture abstraction — registration, bridged media (plain, SDES, codec-mismatch transcoding), byte-exact content, RTCP, hold/unhold, attended transfer, DTMF and a concurrent-call soak (trait InteropFreeSwitch, local-first — not in the PR gate). The Asterisk-only scenarios are listed on the FreeSWITCH page.

Browser + TURN coverage. A dedicated browser-interop suite drives Chromium and Firefox headless via Playwright through the full path — signalling → ICE → DTLS-SRTP → SRTP — with bidirectional Opus and browser-decoded VP8, in both roles (SDK as offerer and as answerer). TURN relay allocation and media are verified end-to-end against a real coturn server.

Soak coverage. The soak suite runs the media-quality matrix (PCMU/Opus × plain/SRTP) plus resource plateau/leak guards over long runs (SoakShort on PRs, SoakLong nightly), asserting no jitter-buffer overflow, correct loss accounting, and no monotone resource drift.

The full test matrix and the L0–L4 test model are documented in CONTRIBUTING.md and MAINTAINING.md. The capacity profile, quality gates, report fields and interpretation limits are documented in docs/maintainers/capacity-quality-benchmark.md.

Quickstart

1. Connect and place a call

using Microsoft.Extensions.Logging;
using CalloraVoipSdk.Core.Domain.Calls;
using CalloraVoipSdk.Core.Domain.Lines;
using CalloraVoipSdk;

using var loggerFactory = LoggerFactory.Create(b => b
    .AddConsole()
    .SetMinimumLevel(LogLevel.Information));

using var client = new VoipClient(new VoipConfiguration
{
    LoggerFactory = loggerFactory,
    UserAgent = "MySoftphone/1.0",
    MaxConcurrentCallsPerLine = 4
});

var connectResult = await client.ConnectAsync(
    new SipAccount
    {
        Username = "1001",
        Password = "secret",
        SipServer = "pbx.example.com",
        DisplayName = "Agent 1001",
        Transport = SipTransport.Tls
    },
    new ConnectOptions
    {
        Timeout = TimeSpan.FromSeconds(15),
        FailFastOnRegistrationFailed = true
    });

if (!connectResult.IsSuccess || connectResult.Line is null)
    throw new InvalidOperationException($"Connect failed: {connectResult.Status}");

var line = connectResult.Line;

var dialResult = await client.DialAndWaitUntilConnectedAsync(
    line,
    "sip:1002@pbx.example.com",
    new DialWaitOptions
    {
        ConnectTimeout = TimeSpan.FromSeconds(30),
        HangupOnTimeout = true,
        HangupOnCancellation = true
    });

if (!dialResult.IsSuccess || dialResult.Call is null)
    throw new InvalidOperationException($"Dial failed: {dialResult.Status}");

var call = dialResult.Call;

await client.AttachDefaultAudioAsync(call);

await call.SendDtmfAsync(new DtmfTone('5'));
await call.HoldAsync();
await call.UnholdAsync();
await call.HangupAsync();

2. Runtime audio device control

var inDevices = client.GetAvailableInputAudioDevices();
var outDevices = client.GetAvailableOutputAudioDevices();

if (inDevices.Count > 1)
    client.SwitchAudioInputDevice(inDevices[1].Id);

if (outDevices.Count > 1)
    client.SwitchAudioOutputDevice(outDevices[1].Id);

client.SetAudioInputVolume(0.8f);
client.SetAudioOutputVolume(1.1f);
client.SetAudioInputMuted(false);
client.SetAudioOutputMuted(false);

client.UpdateAudioFormat(new AudioDeviceFormat
{
    SampleRate = 16000,
    BitsPerSample = 16,
    Channels = 1
});

3. Handle inbound calls

using var subscription = client.OnIncomingCall(async call =>
{
    Console.WriteLine($"Inbound from: {call.RemoteParty}");

    if (IsInLunchBreak())
    {
        await call.RejectAsync(486, "Busy Here");
        return;
    }

    if (ShouldForwardToQueue(call.RemoteParty))
    {
        var result = await call.RedirectAsync(["sip:queue@pbx.example.com"], statusCode: 302);
        Console.WriteLine($"Redirect: {result.Status}");
        return;
    }

    await call.AcceptAsync();
    await client.AttachDefaultAudioAsync(call);
});

static bool IsInLunchBreak() => false;
static bool ShouldForwardToQueue(string remoteParty) => false;

4. Advanced event-driven flow

var line = client.Lines.Register(account);

line.StateChanged += (_, e) =>
    Console.WriteLine($"Line: {e.OldState} -> {e.NewState}");

var call = await line.DialAsync("sip:1002@pbx.example.com");

call.StateChanged += (_, e) =>
    Console.WriteLine($"Call {e.Call.CallId}: {e.OldState} -> {e.NewState}");

5. Manual media control

using CalloraVoipSdk.Audio.Linux;
using CalloraVoipSdk.Core.Application.Ports.Audio;
using CalloraVoipSdk.Core.Domain.Calls;

using var audioDevice = new LinuxAudioDevice();
using var receiver = client.Media.CreateReceiver();
using var sender = client.Media.CreateSender();

call.StateChanged += (_, e) =>
{
    if (e.NewState == CallState.Connected)
    {
        receiver.AttachToCall(call);
        sender.AttachToCall(call);

        var audioParameters = call.MediaParameters is { } mp
            ? AudioConnectionParameters.From(mp)
            : AudioConnectionParameters.Default;

        audioDevice.Connect(receiver, sender, audioParameters);

        if (audioDevice is IAudioDeviceRuntimeControl runtime)
        {
            runtime.SetInputMuted(false);
            runtime.SetOutputMuted(false);
            runtime.SetInputVolume(0.9f);
            runtime.SetOutputVolume(1.0f);
        }
    }

    if (e.NewState == CallState.Terminated)
    {
        audioDevice.Disconnect();
        receiver.Detach();
        sender.Detach();
    }
};

6. Bridge two active calls

using var aRx = client.Media.CreateReceiver();
using var aTx = client.Media.CreateSender();
using var bRx = client.Media.CreateReceiver();
using var bTx = client.Media.CreateSender();

aRx.AttachToCall(callA);
aTx.AttachToCall(callA);
bRx.AttachToCall(callB);
bTx.AttachToCall(callB);

using var bridge = client.Media.CreateConnector().CrossConnect(aRx, aTx, bRx, bTx);

7. Pin the audio codec

using var client = new VoipClient(new VoipConfiguration
{
    UserAgent = "MyVoiceBot/1.0",
    // Order = preference. Offers/answers only include the listed codecs (plus DTMF
    // telephone-event), and RTP sessions pick their primary codec accordingly.
    // Useful for passthrough scenarios, e.g. G.711 µ-law towards a realtime AI API.
    PreferredAudioCodecs = ["PCMU"]
});

8. Video call (bring your own codec)

Video is transport-only — the SDK moves encoded frames but never encodes or decodes. Attach a receiver/sender to a call and drive your own VP8/H.264 codec. The SDK hands you a ready-to-use recommended bitrate and surfaces peer keyframe requests.

using CalloraVoipSdk.Core.Application.Media;

using var videoIn = client.Media.CreateVideoReceiver();
using var videoOut = client.Media.CreateVideoSender();
videoIn.AttachToCall(call);
videoOut.AttachToCall(call);

// Inbound: decode encoded frames yourself (handler runs on the media path — never block).
videoIn.FrameReceived += (_, e) => myDecoder.Decode(e.Frame.Payload);

// The payoff: let the SDK size your encoder to the network.
videoOut.RecommendedBitrateChanged += (_, e) => encoder.SetBitrate(e.RecommendedBitrateBps);
videoOut.KeyFrameRequested += (_, _) => encoder.ForceKeyFrame();

// Outbound: send already-encoded frames.
await videoOut.SendAsync(new VideoFrame(encodedBytes, PayloadType: 96, RtpTimestamp: ts, IsKeyFrame: false));

Prefer the "audio-simple" path? Package your codec behind an IVideoDevice, register it in DI, and call await client.AttachDefaultVideoAsync(call). Full walkthrough: Video calls guide.

Extending the SDK — module registry

client.Modules is the extension point for feature modules that ship as separate packages. A module implements IVoipClientModule, gets attached to the client and is then resolvable by any interface it implements:

// Register (or inject via DI as IVoipClientModule before AddCalloraVoip):
client.Modules.Register(new MyRecordingModule());

// Resolve anywhere:
var recording = client.Modules.Get<IMyRecordingFeature>();      // throws if unavailable
if (client.Modules.TryGet<IMyRecordingFeature>(out var feature)) // or probe
    feature.Start();

Modules build on the public per-call media tap. Its contract in two sentences: IMediaReceiver.FrameReceived fires synchronously on the media path — handlers must buffer and return immediately, never block. Negotiated format details (payload type, clock rate, samples per packet) are available via ICall.MediaParameters.

Commercial plugins (private, paid — in development)

On top of this extension point we are building a set of commercial plugins, distributed through a private feed (not on nuget.org):

• Callora.Realtime — bridge call audio to realtime AI APIs (e.g. OpenAI Realtime) with pacing, backpressure and barge-in support; powers AI voice agents
• Callora.WebSocket — raw call-audio streaming over WebSocket
• Callora.Privacy / Callora.Risk / Callora.Intelligence — redaction & consent, spam/scam screening, AMD/transcription/sentiment

The SDK core stays open and free; plugins are licensed separately. Contact info@bechstein.digital for early access.

Production guidance

• Dispose and unregister VoipClient, IPhoneLine and ICall cleanly
• Execute call actions only in valid states such as Connected, Ringing or OnHold
• Keep event handlers short and non-blocking under load
• Choose audio providers explicitly via platform-specific packages
• Treat infrastructure details as non-public integration surface
• ICE (RFC 8445 / RFC 7675) is opt-in (IceConfiguration.Enabled defaults to false) and, while largely implemented, remains unproven in production — validate it for your trunk before enabling it. The production-proven NAT path is symmetric RTP (comedia), which needs no ICE or STUN.

Roadmap

• Full ICE (RFC 8445 / RFC 7675) is implemented and opt-in — role + tie-breaker, check-list FSM, USE-CANDIDATE nomination, inbound/triggered checks, restart detection and local restart initiation (ICall.RestartIceAsync, RFC 8445 §9 — new credentials on the existing socket, role preserved), and consent freshness with media cease. Final state and selected pair are observable via ICall.IceSnapshot; post-establishment changes (incl. consent loss → Disconnected) via ICall.IceConnectionStateChanged. Remaining gaps toward production (#62): live interop/production validation of the ICE path, and ICE-TCP candidates (RFC 6544) — a deliberate post-GA limitation, since real trunk calls run over symmetric RTP (comedia), which needs no ICE or STUN.
• Commercial plugin line-up (private feed, licensed): Callora.Realtime, WebSocket streaming, Privacy/Risk/Intelligence — in development
• CI/CD hardening: soak, Asterisk interop, chaos/fault-injection, and performance gates are all in place (media-quality matrix, resource-leak guards, full SIP/RTP flow against a real container with zero skipped cases, a two-leg byte-exact bidirectional media test, a per-PR chaos gate that injects transport loss, malformed/adversarial packets, signaling outage, and resource churn under fault and asserts graceful degradation + recovery + no leak, and a per-PR performance gate that holds the SRTP per-packet crypto hot path above a catastrophic-regression throughput floor); the machine-specific capacity envelope runs nightly
• Broader interop validation against more PBXs, trunks and browsers. Automated in CI today: Asterisk (PJSIP), Chromium, Firefox and coturn. FreeSWITCH is automated but local-first; FRITZ!Box is manually verified; Safari/WebKit and the commercial trunks are configuration guidance so far — see the interop matrix

License

Licensed under the Apache License, Version 2.0. See LICENSE.

Contributing

Contributions, issues and interop reports are welcome — this is our first open-source release, so real-world feedback (which PBX/trunk/browser you tested against, what broke) is especially valuable.

• Start here: CONTRIBUTING.md — setup, test levels, and the bar for a PR.
• Report a bug / request a feature: open an issue (templates provided). For larger changes, open an issue first so architecture and API impact can be discussed.
• Good first issues: browse the good first issue label.
• Maintainer docs: MAINTAINING.md, ENGINEERING_RULES.md and docs/maintainers/.
• Code of Conduct: CODE_OF_CONDUCT.md.

Security

Please report security vulnerabilities privately — see SECURITY.md — and not as a public issue. This SDK handles SIP digest authentication and SRTP/DTLS keying, so responsible disclosure matters.

Support

The SDK core is open and free (Apache-2.0). If it helps you build something, you can support ongoing development on Ko-fi ☕. For commercial plugins or early access, contact info@bechstein.digital.