The ElevenLabs voice agent in Amazon Connect.
From US$300 a session: an ElevenLabs voice agent that answers on your own Connect number and hands over warm. It speaks with a natural, human-sounding voice, holds a real conversation with the caller, drives your back-end systems to complete the request, then hands off to a human queue where needed — with the AI summary, caller intent and transcript already in the agent's CRM screen-pop. ElevenLabs conversational voice for the contact centre, wired into Amazon Connect by ProUCX.
Available on AWS MarketplaceElevenLabs Voice Agent in Amazon Connect — by LiveUCX
ElevenLabs voice on Amazon Connect · Live transcript & sentiment · Back-end process automation · Warm handoff with context · Runs against your own Connect instance
For organisations already running Amazon Connect — the ElevenLabs voice works from your existing Connect instance, on the number your customers already call. AWS Connect and Amazon Connect are the same platform, so if you came looking for an ElevenLabs AWS Connect agent, this is it.
The summary, caller intent and transcript travel with the call into the agent's Dynamics 365, HubSpot or Pipedrive screen-pop — the differentiator that most voicebots miss.
A human-sounding voice that holds a free-flowing conversation, so the caller talks normally instead of navigating a press-1 menu.
Answers in your contact flow and hands off as an ordinary Connect transfer, so routing, recording and reporting stay with the contact.
An ElevenLabs voice that hands off warm — not into a void
Plenty of tools can put an ElevenLabs voice on a phone line — enough for a contact center to replace a rigid, menu-driven IVR with an ElevenLabs voice the caller can simply talk to. The hard part is what happens when the caller needs a person. Most voicebots dump the caller into a queue with nothing to show for the conversation, so the human starts from zero and the caller repeats themselves. This integration is built around the handoff: the ElevenLabs agent answers, understands and qualifies, and when a person is needed it passes the call on warm — with the context attached. If you're weighing this against the Amazon Polly voices Connect ships with, we walk through the difference in can Amazon Connect use ElevenLabs voices? — and compare the two engines feature by feature in Amazon Polly vs ElevenLabs.
“Hi, I'd like to change the delivery address on order 4471.”
Summary: wants to change delivery address on order #4471. Intent: order_amend · Transcript: view.
The agent opens the call already knowing what the caller wants.
From an ElevenLabs-answered call to a human, already in context
Your Amazon Connect contact flow routes the incoming call to the ElevenLabs voice agent, which greets the caller and starts a natural conversation — no rigid touch-tone menu.
The agent works out the reason for the call, asks the right follow-ups and captures the details it needs, with the ElevenLabs voice carrying the conversation.
Simple enquiries can be handled end to end. When a human is the right answer, the agent hands off — and the caller can ask for a person at any time.
A summary, the detected intent and a transcript link are written to Amazon Connect contact attributes on the same contact.
The call transfers to the right queue, and the context appears in the agent's CRM the moment they connect.
ElevenLabs voice answers the call in the Amazon Connect contact flow; natural and conversational, not a touch-tone IVR; triages and qualifies before a human is involved; warm handoff to a human queue with the context attached; summary, intent and transcript link written to contact attributes; fail-safe — if the AI can't help, the caller goes straight to a person; works from your existing Amazon Connect instance.
The ElevenLabs conversation shows up where your agent is already looking
The handoff travels through Amazon Connect contact attributes — the same mechanism the ProUCX CRM connectors already read for screen-pop. So the ElevenLabs agent's work surfaces right beside the matching record, with no connector changes.
The summary, intent and transcript link surface in the agent panel when the call screen-pops the matching record. AWS Connect for Dynamics 365
Context lands beside the matching HubSpot record as the call connects, so the rep opens the conversation informed. AWS Connect for HubSpot
The summary and transcript appear against the matching person or deal the moment the softphone screen-pops. AWS Connect for Pipedrive
What the ElevenLabs integration is being built to do
An ElevenLabs voice agent that picks up the Amazon Connect voice contact directly in your contact flow, before an agent is engaged.
Human-sounding, free-form conversation with an ElevenLabs voice — not a scripted DTMF menu.
Set the greeting, prompt and voice so the agent sounds like your brand rather than a generic bot.
Works out why the caller is ringing and routes accordingly — resolve, direct to the right queue, or escalate.
Transfers to a human queue with the AI summary, intent and transcript attached to the contact.
Writes a concise summary and a link to the full transcript to Amazon Connect contact attributes for the CRM to render.
Ask for an agent, stay silent, or hit a limit and the call goes straight to a human — the AI is an optimisation, never a trap.
Designed to surface context in the ProUCX connectors for Dynamics 365, HubSpot and Pipedrive with no connector changes.
Built for Amazon Connect and designed to run against your own AWS account and Connect instance.
Where the ElevenLabs voice is processed
We won't pretend otherwise: the AI voice processing runs on ElevenLabs, which processes audio outside Australia — there is no ElevenLabs region in Australia. Your Amazon Connect instance, call recordings and reporting stay in your own AWS account and region, exactly as they do today. If you have a contractual requirement for the voice AI itself to be processed onshore in Australia, tell us before you commit — that's a conversation to have, and there are onshore-capable paths we can discuss under the broader Live AI Agent product.
Hear what it sounds like, and watch what it does
The demo console runs a call end to end: the lifecycle moving from dial in through to wrap-up, the transcript building turn by turn with sentiment on each message, the back-end integrations lighting up as the agent calls them, and the process-flow checklist completing.
Amazon Connect can now use an ElevenLabs voice natively. Do you still need this?
Amazon Connect’s Set voice block now offers third-party voice providers, ElevenLabs among them. You store the API key in AWS Secrets Manager, enter a model and a voice, publish the flow, and your prompts are spoken in that voice. If a better-sounding prompt is the whole of what you want, that is the whole of the job — and you do not need us for it. The honest version of where our product starts is below.
A Voice provider dropdown on the Set voice block, with fields for the model, the voice, the Secrets Manager ARN holding your API key and the language. At runtime Connect sends your prompt text to ElevenLabs and plays the returned audio to the caller.
It reads what it is handed. It is not listening to the caller, not deciding anything and not looking anything up. The flow still has to know what to say next, which means Amazon Lex in front of it or a human after it. Swapping the voice does not change what the call can do.
Native third-party text-to-speech changes how your contact centre sounds. A conversational voice agent changes what the call can resolve before a person is involved. They sit at different layers and they are not alternatives to one another.
Use the Set voice block. It is a configuration change in a flow you already own, there is nothing to buy from us, and we would rather say so on this page than have you find out after a procurement cycle.
Something that answers, understands why the caller rang, asks the follow-up question, checks the systems that hold the answer and either resolves the call or hands it to a person with the context already attached.
The native voice on your flow prompts and a voice agent on the queue or the hours you choose are not in conflict. Most teams that end up with an agent started by changing the voice first.
The configuration steps are in the AWS documentation for third-party text-to-speech providers. For the sound of the two engines side by side, see Amazon Polly vs ElevenLabs.
There is a native route for the whole agent now, not just the voice
The section above is about the voice, and it still holds. But the line has moved since: ElevenLabs and AWS now document connecting an ElevenLabs agent to Amazon Connect directly, through Connect’s third-party AI agent protocol — an extension of the open A2A protocol. Connect keeps the telephony and the routing; ElevenLabs runs the conversation. No SIP trunk, no third-party carrier, no middleware of ours in the path. If you have read ElevenLabs’ own integration page and arrived here asking why you would need anything else, that is a fair question and this is the answer to it.
A Get customer input block points at an Amazon Lex V2 bot carrying the AMAZON.QInConnectIntent. An Amazon Q in Connect orchestration agent hands the conversation to ElevenLabs, registered through AppIntegrations as an A2A_SERVER application. Audio streams over a WebSocket as 16-bit linear PCM at 8, 16 or 24 kHz. The turn ends when the agent returns Complete or Escalate, which the flow reads back from the Lex session attributes.
A Connect instance on the Customer tier with third-party AI agent support switched on by AWS — ElevenLabs describes this as limited availability and notes it may incur additional AWS charges. Then an Amazon Q in Connect assistant, permissions spanning KMS, Secrets Manager, AppIntegrations and Lex V2, an ElevenLabs workspace with the Amazon Connect transport, and AWS CLI v2 plus awscurl for the calls that released CLI versions do not carry yet.
ElevenLabs lists the limits on that same page. Custom contact attributes set in your flow are not passed to the agent. Data collection results are not returned to the flow. On escalation, Connect receives the outcome but not the reason — their suggested workaround is a webhook that writes a summary back with UpdateContactAttributes, plus a whisper flow to read it out. Client tools and the keypad / DTMF tool are unsupported, and the agent cannot dial a number or a SIP URI from a Connect call.
Read those three limits together and they are one limit: context does not cross the boundary. Nothing of the call goes in, and nothing but a verdict comes out. That is survivable if the agent’s job is self-contained — answer, resolve, hang up — and if that is your job, use the native route and keep the money. It is not survivable when the call ends up with a person, because that person inherits a caller who has already explained themselves once to something that kept no record they can see. That seam is the whole subject of this page, and what we are building is being built to sit on it: the conversation, the reason it escalated and the summary landing on the record your agent already has open, in the CRM they already work in. We would rather set the boundary out plainly than have you discover it in week three of a build.
Where the ElevenLabs audio is actually processed
This is the question that decides the project for Australian government, health and financial-services buyers, so it is worth being precise rather than reassuring. The answer is documented by AWS, and it is not what an onshore-only policy wants to hear.
Unless you say otherwise, Amazon Connect talks to api.elevenlabs.io. You can pin a different provider region by storing the API key as a JSON object with an apiTokenRegion value alongside it.
Three: us, eu and in. Those are the apiTokenRegion values Amazon Connect supports for ElevenLabs, each with its own endpoint.
Not on that list, and we are not going to imply otherwise. If your requirement is that voice audio is synthesised onshore in Australia, ElevenLabs does not currently meet it, whether you reach it through the native block or through us.
Your Amazon Connect instance does not move and your recordings stay in your own AWS account. What leaves the region is the synthesis request and the returned audio. Whether that is acceptable is a question for your privacy and security people, and it should be asked before the pilot rather than after it.
Amazon Polly’s neural voices include Australian English and are processed in your own region. The voice is less natural than ElevenLabs; the residency position is cleaner. That is a real trade-off and only you can price it.
Including us. “Where is the audio processed, and which region is the API key pinned to?” has a documented answer, and any supplier who answers it vaguely is worth a second question.
AWS publishes the supported endpoints and regions for third-party speech providers. Our own hosting and residency position is on the security page.
ElevenLabs Voice Agent questions
No. ProUCX builds the integration that puts an ElevenLabs voice agent into your Amazon Connect contact flow and hands the conversation context into your CRM. ElevenLabs provides the conversational voice AI; ProUCX connects it to Amazon Connect. ElevenLabs is a trademark of its respective owner.
The ElevenLabs Voice Agent in Amazon Connect is listed on AWS Marketplace, so you can subscribe against the AWS account you already have. You can also book a briefing and we'll run it against your own Amazon Connect instance.
The AI voice processing runs on ElevenLabs, which processes audio outside Australia — there's no ElevenLabs region in Australia. Your Amazon Connect instance, call recordings and reporting stay in your own AWS account and region. If you have a contractual requirement for the voice AI itself to be processed onshore, talk to us before you commit — there are onshore-capable paths we can discuss under Live AI Agent.
An AI summary, the detected intent and a transcript link, written to Amazon Connect contact attributes. Our connectors for Dynamics 365, HubSpot and Pipedrive already read contact attributes for screen-pop, so that context lands in the CRM as the call connects.
Yes. This page covers the ElevenLabs option specifically, chosen for voice quality. The broader Live AI Agent product is designed to be voice-vendor-flexible, including onshore-capable options for buyers with an Australian data-residency requirement.
Yes. Open the live demo and watch a call run end to end — the lifecycle, the live transcript with sentiment, the back-end integrations and the process flow completing.
Yes. The Set voice block in a contact flow offers third-party voice providers including ElevenLabs. You store the API key in AWS Secrets Manager, set the model, voice and language, and publish the flow. Connect then sends your prompt text to ElevenLabs and plays back the returned audio.
Not if you only want your existing prompts spoken in a better voice — use the native block. You need something more only when you want the call handled: a voice that listens, asks follow-up questions, checks your systems and resolves or warm-transfers the call. Native text-to-speech reads what it is handed; it does not decide anything.
No. AWS documents three provider regions for ElevenLabs — us, eu and in — plus the default api.elevenlabs.io endpoint. There is no Australian endpoint, so voice synthesis is processed offshore. Your Connect instance and your recordings stay where they are.
A contact flow you can edit, the provider API key stored in AWS Secrets Manager, a Secrets Manager resource policy that lets Connect retrieve the key, AWS KMS permissions to decrypt it, and the provider's model and voice values.
No, and this is the most common misunderstanding. Text-to-speech changes how the prompts sound, not what the flow can do. A menu read in a natural voice is still a menu. Conversation requires something that listens and decides — Amazon Lex, or a voice agent.
Sold through AWS Marketplace in session bundles — US$300 for one session, US$1,000 for five and US$3,000 for twenty — billed against the AWS account you already have. Check the listing for current bundle sizes and terms. Amazon Connect telephony and service charges are billed separately by AWS at your region’s published rates, as they are for any other call.
Yes. ElevenLabs and AWS document connecting an ElevenLabs agent to Amazon Connect through Connect’s third-party AI agent protocol, an extension of the open A2A protocol, with Connect handling telephony and routing while ElevenLabs runs the conversation. ElevenLabs describes it as limited availability and it has to be enabled on your instance by AWS. If the agent’s job is self-contained, that route is enough on its own.
Per ElevenLabs: a Connect instance on the Customer tier with third-party AI agent support enabled by AWS, an Amazon Q in Connect assistant, a Lex V2 bot carrying the AMAZON.QInConnectIntent, and permissions across KMS, Secrets Manager, AppIntegrations and Lex V2. Setup also uses AWS CLI v2 with awscurl, because some of the calls are not in released CLI versions yet. They note it may incur additional AWS charges.
No — ElevenLabs states that custom contact attributes set in the contact flow are not passed to the agent, and that data collection results are not returned to the flow. On escalation the flow receives the outcome but not the reason; their suggested workaround is a webhook that writes a summary back with UpdateContactAttributes and a whisper flow to read it to the agent. Carrying that context both ways is what our integration is being built to do.
Put an ElevenLabs voice on your Amazon Connect number
Subscribe on AWS Marketplace and bill it to the AWS account you already have, or book a briefing and we'll run it against your own Amazon Connect instance.
Available on AWS MarketplaceElevenLabs Voice Agent in Amazon Connect — by LiveUCX
Part of Live AI Agent. This page covers the ElevenLabs option specifically, chosen for voice quality. It's part of the broader ProUCX Live AI Agent product — an AI voice agent for Amazon Connect that's designed to be voice-vendor-flexible, including onshore-capable options for buyers with an Australian data-residency requirement. Explore Live AI Agent →