WaveKat Voice — the softphone for Mac and Linux that records and transcribes every call — now runs on a SIP engine we wrote from scratch. wavekat-sip is an open-source Rust crate that handles the phone signaling and audio transport itself, with no third-party SIP stack underneath. Every call you place or answer, every hold and transfer, now goes through code we own end to end.
This is another step toward the thing we keep coming back to: giving every small business the voice of a big one. A big phone system is dependable because the people who run it control every layer of it. Owning the SIP engine is us controlling the layer your call actually travels on — so when something needs to be more reliable, or a feature needs to ship, there’s no black box in the way.
What SIP is, and why it’s the hard part
SIP (Session Initiation Protocol) is the language phones speak to set up a call — to register your line with your provider, ring the other side, negotiate which audio codec to use, and tear the call down cleanly at the end. RTP is what carries the actual audio once the call is up. Get SIP subtly wrong and calls drop, one-way audio happens, or a line silently stops receiving calls. It’s the unglamorous, exacting core that everything else in a softphone sits on top of.
Until recently, WaveKat Voice drove its calls through a third-party SIP library. That got us on the phone quickly, which was the right call early on. But a borrowed stack decides for you how a call is modeled, how errors surface, and which features are even reachable — and the moment we started building things like call transfer and HD audio, we were working around its shape instead of with it.
Why we built our own
We rewrote WaveKat Voice’s SIP engine from scratch, as wavekat-sip, for three plain reasons:
- Control. Features like putting a caller on hold, transferring a call, and keeping a long call alive with session timers all live at the SIP layer. Owning that layer means we add them directly, instead of bending someone else’s model to fit.
- Footprint. WaveKat Voice is a lightweight desktop app that’s meant to sit quietly out of your way. A focused, purpose-built engine keeps it small — it carries only the SIP and RTP it actually uses, not a general-purpose stack’s worth of everything else.
- No black box. When a call misbehaves, we can read and fix every line between the button you pressed and the packet on the wire. Nothing about how your calls work is off-limits to us.
What the engine handles
wavekat-sip owns the wire-level concerns and stays out of the audio-device and call-orchestration layers, so it remains small and embeddable:
| Area | What it does |
|---|---|
| Registration | Registers your line with your provider (digest auth), and keeps it alive so incoming calls always reach you. |
| Calls | Places outbound calls and answers inbound ones, and alerts callers with a proper ringing signal before you pick up. |
| In-call control | Hold and resume (SIP re-INVITE, RFC 3264), blind and attended transfer (SIP REFER, RFC 3515), and DTMF (keypad tones) for phone menus. |
| Audio quality | Negotiates the Opus codec for wideband “HD” voice, with automatic fallback to standard G.711 when the other side doesn’t support it. |
| Reliability | RFC 4028 session timers keep long calls from being silently dropped by the network in the middle. |
It’s open source — like the rest of WaveKat
wavekat-sip isn’t a private internal component. It’s published on crates.io under the Apache-2.0 license, with docs on docs.rs, the same way our voice-activity detection and turn detection crates are. Anyone building a softphone, a voice bot, or a call-recording bridge in Rust can use the exact engine WaveKat Voice runs on. Building in the open is how we work — the tools underneath our product are yours to inspect and reuse, not a moat.
It’s honest to say it’s early: the crate is in active development and its API still changes between versions. But it’s the real engine behind a real product, not a demo.
Frequently asked questions
What is wavekat-sip?
wavekat-sip is WaveKat’s own open-source Rust crate for SIP signaling and RTP audio transport. It’s the engine behind every call WaveKat Voice places or answers, with no third-party SIP stack underneath.
Is wavekat-sip open source, and can I use it in my own project?
Yes. wavekat-sip is published on crates.io under the Apache-2.0 license, with docs on docs.rs. Anyone building a softphone, a voice bot, or a call-recording bridge in Rust can use the same engine WaveKat Voice runs on.
Does WaveKat Voice support HD audio?
Yes. WaveKat Voice negotiates the Opus codec for wideband “HD” voice, and falls back automatically to standard G.711 when the other side of the call doesn’t support Opus.
Does WaveKat Voice work with any SIP provider?
Yes. wavekat-sip handles standard SIP registration with digest authentication, so it works with any SIP-compliant provider or PBX — the account you already have, with no provider-specific setup.
Is wavekat-sip production-ready?
It’s the real engine behind a real product, so it’s already in daily use — but it’s early. The crate is in active development and its API still changes between versions, so pin a version if you build on it today.
What this means for your calls
Mostly, WaveKat Voice’s calls won’t feel any different — and that’s the point. They connect and sound the way they should. What changes is behind the scenes: the features that make WaveKat Voice feel like a real front desk — hold, call waiting, transfer, HD audio — now ship on our timeline instead of a dependency’s, and when something needs to be steadier, we can go straight to the code that runs it.
WaveKat Voice is free during the public beta on Mac and Linux. Connect the phone provider you already have, and your next call runs on an engine we built ourselves — and gave away.