Building a voice AI agent has never been easier. LLMs are highly capable, and convincing proof of concepts that can be achieved in days.
The difficulty often comes later: integrating the agent with real telephony infrastructure, maintaining call quality under load, and getting through the security review that tends to arrive only after the pilot has already succeeded.
These are the problems that cause production deployments to stall, and they are the problems AudioCodes Live Hub is designed to address.
How Live Hub Integrates With Existing Contact Centre Infrastructure
“The main idea of Live Hub is to integrate the old things, not to replace them,” says Ilan Avner, Director of Product Management at AudioCodes.
“The idea is to work in parallel with contact centres and enrich them with voice AI that you can get from multiple providers.”
In practice, Live Hub connects to whatever telephony stack is already running and acts as the layer between voice channels and the AI stack. It integrates with tens of conversational AI platforms and tens of ASR and TTS engines. Organisations that have already built bot logic on Google CX Agent Studio, Copilot Studio, or Rasa can connect those directly. Those that have not can build natively within Live Hub. Either way, the existing contact centre infrastructure stays in place.
Keeping the Voice AI Stack Open to Best-of-Breed Providers
Where Live Hub differs from much of the market is in how it handles the underlying technology stack. Many voice AI platforms bundle their own tools, which means the choice of platform is also a choice of provider. Yehuda Herscovici, VP of Product at AudioCodes, explains the alternative:
“If you go to VAPI or Retell, you have to use their own bot-building platform. If you go to ElevenLabs, you must use their own text-to-speech technology. With Live Hub, we do not force our users to use any type of the technology stack. We are committed to being vendor-agnostic, not only now, but also in the future.”
That matters because the market keeps moving. ASR and TTS performance, LLM pricing, latency benchmarks: all of these shift on a short cycle. Being locked into a specific provider at any layer means absorbing those changes rather than responding to them.
Running Voice AI at Scale: Where the Real Engineering Challenge Begins
Most voice AI pilots run at a handful of concurrent sessions. Production deployments don’t, and the gap between the two is where architectures that looked solid can fall apart.
Telephony channels, speech-to-text engines, and LLMs all have concurrency limits that rarely get tested during a pilot, and when they do surface it is usually in front of real customers.
AudioCodes runs deployments at tens of thousands of concurrent sessions, which Herscovici says are among the largest voice AI deployments in the world. At that volume, capacity is only part of the challenge:
“You cannot allow yourself that something will go wrong and thousands or tens of thousands of sessions will be shut down. You need geographical redundancy and seamless switchover so that everything moves to another geography if there is a data centre event.”
Designing for that kind of availability is not something that can be done after the fact: it has to be part of the architecture from the start.
Deployment options for regulated sectors
For regulated industries like banking, insurance, and healthcare, the SaaS model that makes Live Hub fast to adopt is not always viable. Security and governance requirements that were absent during the pilot tend to surface quickly once the wider organisation is involved.
For those environments, AudioCodes offers Voice AI Connect Enterprise: a dedicated deployment installable on-premise or as a private cloud instance, for organisations where data cannot leave their own infrastructure. Some customers are regulated to the point where AudioCodes itself cannot access the deployed systems remotely. Voice AI Connect Enterprise is built for that.
The choice between the two typically comes down to the regulatory environment rather than the use case. Both are built on the same principle: that getting voice AI into production is an infrastructure problem as much as an AI one.
More on both platforms is available at AudioCodes Live Hub and AudioCodes Voice AI Connect.
