Learn / Desk phones
Most guidance about adding calling to a software platform quietly assumes the call happens in a browser tab.
Often it does, and plenty of vendors serve that half well. But it's only half. The front desk at a dental practice has a phone on the counter. The dispatcher at a plumbing company has one with a headset. The service advisor at a dealership has one with six line keys and a directory of extensions. Those same businesses have an owner taking calls on her mobile and a biller working from a laptop, so what they actually need is both, on the same extensions, ringing in the right order.
The hardware half is where the field thins out. Most of the vendors you'll be pointed toward can't help with it at all, and the ones that can differ enormously in how much of the work lands on you.
Physical phones speak SIP. That rules out the browser SDK vendors outright, and turns CPaaS into a build project.
Agora, Daily, LiveKit, Stream, and similar providers build excellent low-latency audio for browsers and mobile apps. They're the right answer for in-app voice chat, telehealth video, and app-to-app calling.
What they can't do is drive a desk phone. A Snom, Yealink, or Poly handset runs vendor firmware that registers over SIP. It has no way to load a JavaScript SDK. If your customers only ever call from inside your app, this category is a reasonable answer, and plenty of vertical AI platforms run their voice agents on exactly these SDKs. The moment a customer wants a handset on a counter, it's out.
Twilio, Telnyx, Bandwidth, Plivo, and SignalWire all support SIP registration, so a physical phone can register to a SIP domain you control and place and receive calls. That part works.
What they don't do is manage the hardware. A CPaaS gives you call control and numbers. It doesn't give you a way to take a phone out of a box at a customer site and have it come up configured. To get there you're building:
This is well-understood work, which is exactly why it's easy to underestimate. It's also permanent. Every new phone model your customers buy is a new template, and every firmware release is a regression risk on hardware sitting in offices you don't control.
A CPaaS also doesn't take the regulatory layer off you, though plenty of people assume it does. If you sell the service to your customer, you're generally the provider regardless of who you buy wholesale from. That's a bigger question than hardware, so it's covered separately in who is legally the carrier when you sell phone service.
Buying SIP trunks direct from a carrier is the cheapest path per minute, and if you're doing serious volume the margin is real. But trunks are raw dial tone. To register phones against them you need a PBX in the middle: FreePBX, Asterisk, FreeSWITCH, Kamailio, or a commercial equivalent.
That means you own multi-tenancy, high availability, media handling, NAT traversal, call recording storage, and the on-call rotation for all of it. You also own the provisioning stack from option 2, because a PBX doesn't remove that work, it just relocates it.
The regulatory layer on top is larger than the engineering one, and it's the same one from option 2: buying wholesale doesn't move the line. It gets its own treatment in who is legally the carrier when you sell phone service.
Some platforms should still do this. If voice is going to be a core product surface and you have telephony expertise in house, it's a defensible choice. Just price the whole thing rather than the engineering half.
Intermedia, Yeastar, NetSapiens, RingLogix, and DialStack all ship device provisioning built in.
That's what the industry calls zero-touch provisioning. You enter a phone's MAC address in an admin portal and assign it to a user. The phone is shipped to the customer, plugged into ethernet, and on boot it reaches out, pulls its configuration, and comes up with the right extension, line keys, and voicemail. Nobody has to type a SIP password into a handset.
Inside this category, four things actually differ: which devices are certified, whose brand the admin portal wears, whether the platform was built for MSPs or for software platforms, and how much of the regulatory layer the vendor takes on.
These separate the real answers from the demos. Ask all of them.
Two hundred customers averaging six phones each is twelve hundred devices in offices you've never visited, on networks you don't control, running firmware you didn't choose. Every one of them is a potential support call that arrives to you, because you're the software vendor whose name is on the invoice.
That's the real question behind vendor selection. Not whether a desk phone can ring, but who's on the hook the day all twelve hundred are fine except one.
An Embedded Phone System and VoiceAI built for vertical SaaS and vertical AI platforms rather than for MSPs and resellers. Desk phones, cordless handsets, and softphones in the browser or on mobile all run on the same extensions. The admin portal carries your brand instead of ours, so your customers never see a second vendor.
We're not arguing for hardware over softphones. Most of your customers will want a mix, and the mix is the point.
That part is our job, permanently.
Explore the device guide ยท Migrating existing phones
Phones are one part of this. What makes a platform embed a phone system instead of referring one out is the call itself: the customer record opens on the first ring, the call logs itself, and a voice agent can answer, schedule, and summarize. See what that looks like.
Bring an engineer. We'll walk your architecture and exactly where desk phones drop into your product. If the answer is that you should build it on a CPaaS, we'll tell you that.