Custom AI voice agents on ElevenLabs

Custom AI voice agents on ElevenLabs

We build custom AI voice agents on ElevenLabs. Sometimes that’s a complete ElevenLabs voice agent on ElevenAgents, the agent platform also known as ElevenLabs Agents, and sometimes it’s an agent on Vapi, Retell AI, Synthflow or the AgentVoice engine that speaks with an ElevenLabs voice. ElevenLabs voices go into our builds on every platform we work with except Bland, where agents use Bland’s own voices.

What ElevenAgents is

ElevenAgents is ElevenLabs’ platform for configuring, deploying and monitoring agents that talk over voice, chat and email. ElevenLabs’ docs describe each agent as a speech-to-text model, a language model you choose, ElevenLabs’ own text-to-speech and “a proprietary turn-taking model,” which is the part that decides when the caller has finished and the agent should reply. The pricing page and docs use the ElevenAgents name, while some older doc links still point to an agents-platform path and the ElevenLabs site footer still links to Conversational AI.

You can build agents in the dashboard, where a visual Workflows editor lays out branching conversations, or manage them as code with the ElevenLabs CLI, SDKs for React, Swift, Kotlin and React Native, and a WebSocket API. For the language model, you can use ElevenLabs-hosted Qwen, Google Gemini, Anthropic Claude or OpenAI GPT, bring your own OpenAI key, or point the agent at a custom LLM server, with cascading fallback between models.

ElevenLabs sells ElevenAgents on self-serve plans that start with a free tier, plus an Enterprise tier. Enterprise clients get forward-deployed engineers, who ElevenLabs says help them “take their first agents to production within 2-3 months.”

ElevenLabs Agents pricing

As of September 2026, ElevenAgents plans are billed by call minutes, separate from the credit pool ElevenLabs uses for its other products. Each plan includes a set number of minutes and concurrent calls, according to the ElevenAgents pricing page.

Plan Monthly price Call minutes included Concurrent calls
Free $0 15 4
Starter $6 75 6
Creator $22 ($11 the first month) 275 10
Pro $99 1,238 20
Scale $299, 3 seats 3,738 30
Business $990, 10 seats 12,375 40
Enterprise Custom Custom Elevated limits

Minutes past the allowance cost $0.080 each on every plan, and that’s also what the included minutes work out to. Pro’s $99 for 1,238 minutes and Business’s $990 for 12,375 minutes both come to $0.080 a minute, so on a paid plan you’re prepaying minutes at a flat rate, and the plan you choose mainly sets concurrency and seats. Annual billing is priced at 10 times the monthly price.

The language model is billed on top, based on the model you choose, and the calculator shows Gemini 2.5 Flash at $0.0012 a minute as one example. ElevenLabs adds no telephony fee, so your Twilio or SIP provider bills you directly. Calls above your concurrency limit can run as burst calls at $0.160 a minute, up to three times your limit or 300, whichever is lower, and ElevenLabs notes those calls are deprioritized and may see higher latency. Its help center also describes a 95% discount on stretches of silence longer than 10 seconds.

Healthcare work needs Enterprise. A BAA is only available on the Enterprise tier, and calls involving protected health information have to run in Zero Retention Mode, which limits the agent to an allowlist of language models and rules out MCP tools. Data residency is also an Enterprise feature.

What we build on ElevenLabs Agents

An ElevenAgents build runs like the rest of our AI voice agent development work. We scope the calls, design the Workflows graph and prompts, connect the tools the agent uses mid-conversation, and test on real calls before launch. The tools we set up cover:

  • Webhook tools that look up and update records in your CRM, calendar or help desk
  • Transfers to a staff phone number or to another agent
  • Voicemail detection on outbound calls, so the agent leaves a message or ends the call
  • Keypad tones to get through phone trees, and keypad input from callers for account numbers or menu choices
  • A knowledge base with retrieval, planned around ElevenLabs' note that RAG adds about 250ms to a response
  • Language detection so the agent can answer in the caller's language
  • Batch outbound calls from a CSV or spreadsheet, scheduled on Twilio or SIP trunk numbers

Transfers depend on how the number is connected. Conference transfer is the default, blind transfer works only on native Twilio numbers, SIP REFER is supported, and warm transfer messages need a number imported through the native Twilio integration, so when staff expect a warm handoff we connect the line through Twilio. We can also manage the agent as code with the ElevenLabs CLI, which puts prompt and tool changes under version control.

ElevenLabs voices on other platforms

When the agent runs on another platform, we still use ElevenLabs voices, and each platform bills them its own way. Vapi lists ElevenLabs among 18 voice providers and lets you connect your own ElevenLabs key, so ElevenLabs bills you directly for the voice. Retell AI hosts ElevenLabs voices itself at $0.040 a minute, against $0.015 for its other voices, and doesn’t document using your own key.

Synthflow uses ElevenLabs as its default voice engine, with Flash v2 for English and Flash v2.5 for multilingual calls, and accepts your own ElevenLabs key. On the AgentVoice engine in your own cloud, ElevenLabs is the voice we build with. Bland is the exception, since it runs its own voice engine and we found no ElevenLabs option in its docs.

When another platform fits better

ElevenAgents suits clients who want ElevenLabs’ voices and turn-taking with published per-minute rates and a plan small enough to pilot on. A few requirements send us elsewhere.

HIPAA without an Enterprise contract is one of them. ElevenLabs signs BAAs only on Enterprise, while Retell AI’s BAA is self-signed at no additional fee and Vapi sells HIPAA mode at $2,000 a month per organization. Our Vapi vs Retell page compares those two in detail.

Data residency on a smaller budget is another. Because residency is an Enterprise feature on ElevenLabs, a team with EU callers and modest volume may do better on Vapi’s separate EU region, which we cover on our Vapi pricing page.

Ownership is the last. ElevenLabs offers private deployments that run ElevenAgents inside your own AWS or Google Cloud account, and it says an Azure version is planned for later in the second half of 2026. If you’re on Azure now, or you want the source code as well as the deployment, we’d put the AgentVoice engine in your Azure or AWS account or on your own servers under our platform license. Records, transcripts and recordings then sit in your database under your retention policy, and cloud and AI costs are billed at cost with no per-minute platform fee.

Frequently asked questions

How much do ElevenLabs Agents cost?

As of September 2026, ElevenAgents plans run from free, with 15 call minutes a month, to $990 a month for Business, with 12,375 minutes, and Enterprise is priced on request. Extra minutes are $0.080 each on every plan, the language model is billed on top by model, and your phone provider bills telephony directly.

Is ElevenAgents the same as ElevenLabs Conversational AI?

ElevenAgents is the name ElevenLabs now uses on its pricing page and in its docs for its agent platform. Some older ElevenLabs doc links still point to an agents-platform path, and the ElevenLabs site footer still links to Conversational AI.

Can an ElevenLabs voice agent answer phone calls?

Yes. ElevenAgents connects to phone numbers through a native Twilio integration or SIP trunking, with documented connections for Telnyx, Plivo, Exotel, Bandwidth, Vonage, Five9, Microsoft Teams and WhatsApp. ElevenLabs adds no telephony fee, so your phone provider bills you directly.

Can ElevenLabs Agents run in our own cloud?

ElevenLabs offers private deployments that run ElevenAgents in your own AWS or Google Cloud account, arranged through its sales team under NDA, and says an Azure version is planned for later in the second half of 2026. We can also deploy the AgentVoice engine into your Azure or AWS account or onto your own servers under our platform license.

Is ElevenLabs HIPAA compliant?

ElevenLabs' compliance portal lists HIPAA along with SOC 2 Type 2 and ISO 27001. A BAA is only available on Enterprise, and calls with protected health information have to use Zero Retention Mode, which restricts the language models an agent can use.

Can we use ElevenLabs voices on Vapi or Retell?

Yes. On Vapi you can connect your own ElevenLabs key and ElevenLabs bills you directly. Retell hosts ElevenLabs voices itself and charges $0.040 a minute for them, as of September 2026.