One live call, whatever protocol each party is on.
The Xebit audio bridge joins SIP, WebRTC, WebSocket and gRPC in a single live call. While an AI agent talks to a customer on the phone, a person on your team can join from a browser to whisper, join in or take over, with no transfer and no redial. One recording covers every leg.
Voice runs on three islands. Calls need all of them.
Customers call on SIP
The phone network reaches you over SIP, through your own carrier or SBC.
Your team works on WebRTC
Supervisors, specialists and agents live in a browser or an app, not on a desk phone.
AI speaks WebSocket and gRPC
Agents, models and your own services stream audio and events over WebSocket and gRPC.
Every party keeps its protocol. The bridge makes it one call.
| Leg | Protocol | Who is on it |
|---|---|---|
| The customer's phone | SIP (RTP over SIP), through your carrier or SBC | Callers, inbound and outbound |
| A browser or an app | WebRTC | Your team, or customers calling from your app |
| Your own services | WebSocket | Audio streamed to your services or third party bots |
| Internal services | gRPC | AI agents and services exchanging audio and call events |
Relay and fan out
Audio and call events move between every party on the call in real time, whatever protocol each one arrived on.
Codec transcoding
Telephony and WebRTC codecs, such as G.711 and Opus, are converted automatically. Nobody configures it per call.
One recording, every leg
One recording and one transcript of the whole call, including the parts a person joined from a browser.
Agent to agent handoff
One AI agent can pass the live call to another agent on a different service, without the caller noticing a seam.
A person can step into any call. Three ways.
Whisper
- Who hears whom
- The person speaks to one side only. The customer does not hear them.
- Use it when
- Coaching a colleague, or steering the conversation without interrupting it.
Join
- Who hears whom
- Everyone hears everyone: the customer, the AI agent and the person.
- Use it when
- A specialist answers one question, or a team lead confirms an exception, and the agent carries on.
Take over
- Who hears whom
- The person continues the call. The AI agent drops off or goes quiet.
- Use it when
- Disputes, distress, compliance sensitive moments, or anything the agent should not handle alone.
Why this changes how voice agents get built.
| Without a bridge | With the Xebit audio bridge | |
|---|---|---|
| Escalating to a person | Transfer to another system, often a redial, and the caller explains again | A person joins the same call from a browser |
| Coaching | After the fact, from recordings | Whisper during the call |
| Recording and transcript | One per system, stitched together later | One recording and transcript across every leg |
| Adding a channel | A new stack for each protocol | A new leg on the same bridge |
| Several specialised agents | Bot to bot transfers that drop context | Agent to agent handoff on the same call |
Useful on day one, for everyone on the call.
For enterprises
- A team lead steps into a collections call the moment a borrower disputes an amount, from their browser.
- Sensitive moments go to a person on the same recorded call, so the audit trail has no gaps.
- Keep your carrier and your existing setup. Your team joins from wherever they work.
For your customers
- No hold music, no redial and no new number to call when they need a person.
- The call carries on in the same conversation when someone joins.
- A person when the moment needs one, and the agent the rest of the time.
For teams building voice agents
- Build the agent once. Callers can arrive by phone, browser or app.
- Stream audio to your own models and services over WebSocket or gRPC.
- Split work across specialised agents that hand calls to each other.
Runs wherever the platform runs.
- Xebit VoiceCloudManaged by us, priced per connected minute.
- On premInside your network, next to your systems.
- Air gappedInside your network with models on your GPUs. Nothing leaves the network.
Connecting voice agents across protocols, on any stack.
- SIP · WEBRTC · WS · GRPCWhat is a multi-protocol audio bridge, and why do voice agents need one?Read
- WHISPER · JOIN · TAKE OVERHow a human joins an AI agent's phone call from the browser: whisper, join or take overRead
- WARM TRANSFER · SAME CALLAI to human handoff without a redial: why warm transfer on the same call mattersRead
- FOUR PROTOCOLS · ONE CALLSIP, WebRTC, WebSocket or gRPC: which protocol for which leg of a voice agent call?Read
- 8 KHZ · 48 KHZG.711, Opus and transcoding: what happens to audio between the phone network and your AIRead
- ONE CALL · ONE RECORDOne recording across every leg: call recording and transcripts when parties use different protocolsRead
- AGENT TO AGENT · LIVE CALLAgent to agent handoff: passing a live call between specialised AI agentsRead
- EDGE · BRIDGE · AGENTBuilding voice agents without protocol lock-in: an architecture guide for startupsRead
Audio bridge questions
What is a multi-protocol audio bridge?
A multi-protocol audio bridge joins parties that use different real time protocols into one live call. The Xebit bridge relays audio and call events between SIP, WebRTC, WebSocket and gRPC, transcodes codecs such as G.711 and Opus, and records the whole call once, so a phone caller, an AI agent and a person in a browser can share a single conversation.
Can a person join an AI agent's phone call from a browser?
Yes. While an AI agent talks to a customer over SIP, a person on your team can join from a browser over WebRTC. They can whisper to one side without the customer hearing, join as a third party so everyone hears everyone, or take over the call from the agent. The customer stays on the same call throughout.
Does Xebit provide the phone line?
No. Xebit never provides telephony or numbers. The SIP side of the bridge connects to the carrier or SBC you already use, so your numbers, routes and rates stay yours. The bridge adds browsers, apps and your own services to those calls.
Is the whole call recorded when parties use different protocols?
Yes. The bridge produces one recording and one transcript of the whole call, across every leg, whichever protocol each party used. That includes the part where a person joined from a browser, which matters for quality review and audits.
Does the audio bridge run on prem and air gapped?
Yes. The bridge is part of the Xebit platform, so it runs wherever the platform runs: on Xebit VoiceCloud, on your own infrastructure, or fully air gapped inside your network.
Join an AI agent's call from your browser.
We will put an agent on a call, then show you whispering, joining and taking over from a browser, on our cloud or your infrastructure.