SIP and SIP trunking
In one sentence
SIP, the Session Initiation Protocol, is what phone systems use to open, manage and close voice calls across internet networks, and SIP trunking is the service you buy to reach the outside telephone network through an internet connection rather than through physical lines into the building.
Not to be confused with WebRTC, or PSTN.
Definition
SIP is the language phone systems talk when they open, run and close a call over the internet.
SIP trunking means buying that link as a service rather than pulling physical phone lines into a building.
The Session Initiation Protocol, written up as IETF RFC 3261, handles the parts of a call a person would recognize: placing it, making the far end ring, picking up, hanging up. What it never touches is the audio. That goes its own way over RTP, using parameters the two ends settled between themselves via the Session Description Protocol, SDP, which rides inside the SIP messages.
The separation that confuses newcomers
- Control is SIP's job. Who is calling whom, and what should happen next.
- Media is RTP's job. The voice itself travels there.
- Because each takes its own route through the network, each can break on its own. Hence a call that connects and then delivers nothing, the classic one-way audio problem, which nearly always turns out to be the media path and not the signaling.
SIP trunking in commercial terms
- An internet connection to a provider stands in for the physical circuits, ISDN or analog lines, that used to run into the building.
- What you are buying is concurrent channels instead of physical lines, and adding more of them involves nobody visiting the site.
- Pricing usually pairs a capacity or channel fee with per-minute charges on outbound, and inbound is frequently thrown in.
- Numbers stop being tied to geography, so a business can answer on a local number in a city it has never set foot in.
How SIP shows up inside an AI voice agent
- Any agent answering the phone has to get to the phone network somehow, and SIP is almost always how: a trunk of your own, a CPaaS provider hiding the details, or an interconnect into whatever contact center platform is already running.
- Large organizations rarely want the agent to replace the estate; they want it beside the estate. That means interconnecting over SIP rather than issuing a new number, and it is an integration requirement that gets underestimated more often than not.
Practical operational concerns
- NAT traversal, the usual culprit behind one-way audio.
- Timers on the session and the re-invites that go with them, which start to bite on long calls.
- Codec negotiation, since what gets agreed there decides both the audio quality and what transcoding costs.
- Fraud exposure. SIP credentials that leak end up funding toll fraud, and that is an expensive and entirely real risk.
Common misconception
That this is legacy technology on its way out. For business voice it remains the signaling protocol of record, and nothing is displacing it. The change is happening behind it instead: put an AI agent on the far end of a SIP call and you have found a new use for an old and stable protocol.
Why it matters commercially
Where AI voice meets the telephony estate a business already owns, SIP is the seam. Offer a deployment that leaves numbers, carrier and contact center platform exactly as they are and you will close far more of them than a vendor asking for a rip and replace, and interconnecting over SIP is the mechanism that lets you.
In voice specifically
The failure mode has no text equivalent. A messaging integration either delivers a message or it does not. A voice call can be fully connected, correctly signaled and completely silent, because the media took a different road from the control messages.
Where AsqVox fits
Nothing in the Orb touches SIP. This entry is here for buyers who have looked at telephony agents first, hit a wall of SIP requirements, and need to hear that none of it carries over to something running in a browser. On-site, the absence of trunks, numbers to provision and SIP itself is one less thing to build rather than a gap in the product, and installation is still a single script tag.
Visual
Signaling and media take different roads
Two protocols, two failure modes. Silence on a connected call is always the media path.
Statistics
Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.
The protocol is written up in IETF RFC 3261. Open, stable, and implemented just about everywhere.
IETF RFC 3261industry rangeIETF RFC 3261, 2026 - Verifiable from the specification rather than measured. Anchor a citation on the RFC number, not on the year recorded against this row.
Media parameters, codec choice among them, are negotiated by SDP, the Session Description Protocol, which travels inside the SIP messages themselves.
SDP inside SIPindustry rangeIETF SIP and SDP specifications, 2026 - This is the mechanism behind codec negotiation, which is where transcoding cost and audio quality get settled on a call.
Media travels on RTP, which runs independently of the SIP signaling.
RTP carries mediaindustry rangeIETF SIP and RTP specifications, 2026 - The specification fact behind the one-way audio problem. Independent paths mean independent failures.
Trunk capacity is sold by the concurrent channel, and the usual commercial shape puts a capacity charge together with per-minute billing on outbound.
Capacity charge plus per-minuteindustry rangeCommon commercial structure for SIP trunking, 2026 - That shape holds across providers. The rates poured into it do not, and they swing wildly with provider, geography and volume.
There is no dependable benchmark price for a SIP trunk. Between one market and another the rates move by orders of magnitude, so any single figure misleads.
-no reliable figureTake your own expected concurrency and outbound minutes to a provider and ask for a quote. Anything handed to you as the industry price is nothing of the sort.
Neither global SIP trunking penetration nor the speed at which ISDN is being switched off has an authoritative published figure recent enough to cite with confidence.
-no reliable figureNumbers do circulate. They tend to be regional, dated, or drawn from vendor material, and they get repeated as though they were global and current.
Loss figures for toll fraud are quoted constantly across industry material, yet they lead back to aging industry-association estimates rather than to anything measured independently and recently.
-no reliable figureBe wary of any particular number. The exposure is genuine and costly; the headline attached to it usually has no source behind it.
Examples
In practice
An insurer wants to put an AI agent on the claims line it already publishes, keeping both the number and the carrier untouched. The agent hangs off a SIP interconnect into the session border controller that was there before it, picking up calls the carrier is delivering anyway. When a caller needs a person, the same interconnect hands them back into the contact center queue. Nothing is ported, nobody changes carrier, the incumbent telephony supplier keeps its seat, and internal resistance drops accordingly.
The everyday version
What SIP trunking replaced was the bundle of physical lines that used to run into a building. Copper set a hard ceiling on simultaneous calls; an internet connection does not, so you simply buy as many at once as the business needs. SIP itself is the language those calls speak when they announce themselves.
Usage
Who says it
- Telephony engineers, network teams and contact center technologists, fluently.
- Enterprise IT, whenever voice is being procured, and the subject comes up early rather than late.
- Vendors selling AI voice into established businesses, who either speak this language convincingly or watch the incumbent keep the account.
Where it turns up
- Grouped with carrier relationships, porting of numbers, session border controller compatibility, how many concurrent channels are supported, which codecs, and what happens on failover or in a disaster.
- Very often bolted to a demand for integration with some named incumbent platform, and that demand is generally the actual requirement hiding behind the SIP question.
Common misuse
- Reading SIP support as telephony readiness. Getting the signaling to interoperate is necessary and nowhere near sufficient. What sinks deployments is media handling, transcoding and NAT traversal.
- Presuming a browser-based agent has to have SIP behind it. None of it is needed, and stating that outright reads as a strength rather than something left out.
- Swapping SIP and VoIP as though they meant the same. VoIP is the broader thing; SIP is one of the protocols working inside it.
Questions people ask
What is the difference between SIP and RTP?
SIP does the signaling: placing the call, ringing it, answering it, ending it, and carrying no voice at any point. RTP does the media, moving the actual audio on parameters that SDP negotiated inside the SIP messages. Each goes its own way through the network and each can fail alone, which is precisely how a call ends up connected and silent.
Why does a SIP call connect with no audio?
Because the signaling worked and the media path did not. One-way audio is the classic name for it, and NAT traversal is behind it far more often than anything in the SIP exchange. Start with the media path. A connected call is evidence that the control messages arrived, and evidence of nothing else.
Does a website voice agent need SIP trunking?
No. SIP trunking is how a telephony voice agent reaches the outside phone network. A voice agent running in a browser uses browser-native real-time audio, so there is no number to provision, no carrier and no trunk. Buyers who researched telephony agents first often assume the requirement carries over, and it does not.
How much does SIP trunking cost?
No dependable benchmark exists. The usual shape is a channel or capacity charge alongside per-minute billing on outbound, with inbound frequently bundled in, but move between markets, providers or volume tiers and the rates shift by orders of magnitude. Anything handed to you as the industry price is nothing of the sort. Take your own expected concurrency and outbound minutes to a provider and ask.
Last reviewed 30 July 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.