What it is
This page covers ElevenAgents, the voice-and-chat agent platform from ElevenLabs — not the text-to-speech, dubbing, or music products the company is better known for. ElevenAgents lets a team build an agent in a workflow builder or through the API and SDKs, ground it in a knowledge base, pick the LLM, attach a phone number or its own SIP trunk, and deploy it on phone lines, a web widget, and text messaging. It sits in the same build-it-yourself layer as Retell AI and Vapi, with one difference that decides most evaluations: ElevenLabs makes the speech models those platforms resell.
The company is the best-capitalized independent in voice AI. It raised a $500M Series D on 4 February 2026, led by Sequoia at an $11B valuation, and a third close announced 5 May added BlackRock, Wellington, D.E. Shaw, and NVIDIA’s NVentures. It ended 2025 above $330M ARR and said it passed $500M ARR within the first four months of 2026. Named agent customers include Revolut, Klarna, Deutsche Telekom, and Square.
Why it shows up in CX and support stacks
- Voice quality and language coverage are the product, not a line item. Revolut put an ElevenLabs agent on first-line phone support for more than 4 million UK and European customers in January 2026, live in 31+ languages with real-time language switching; the vendor-published case study reports time-to-resolution cut by more than 8x in the first phase. On other platforms ElevenLabs voices are a premium add-on — on Retell they run $0.04–$0.10 per minute on top of the base rate.
- The reference deployments are large and regulated. Klarna made an ElevenAgents voice agent the first point of contact for US phone support in February 2026, reachable by all 35 million US customers, handling payment-status and product questions and handing complex or sensitive calls to humans. Klarna reports resolutions up to 10x faster on the calls the agent handles.
- It connects to the helpdesk, not only the CRM. Documented integrations cover Salesforce, HubSpot, Zendesk (first-party, with ticket lookup, creation, and routing), ServiceNow, Twilio, Vonage, and Genesys, plus any API exposed over MCP.
- Testing is built in. Agents can be run against simulated conversations built from past transcripts or new scenarios before they take a live call — the step most self-built voice projects skip.
Pricing reality
The published ladder runs from Free (15 call minutes) through Starter at $6/month (75 minutes), Creator at $22 (275), Pro at $99 (1,238), Scale at $299 (3,738), and Business at $990 (12,375 minutes, 40 concurrent calls). Every tier works out to the same $0.08 per call minute, and additional minutes bill at $0.08. Telephony carries no ElevenLabs fee; bring Twilio or a SIP trunk and that carrier bills you directly.
The $0.08 is not the all-in rate. LLM usage is charged on top, varies by model, and comes out of the same credit balance. Calls above your concurrency cap run in burst mode — up to 3x the limit — at double rate, $0.16 per minute. Your real rate is $0.08 plus the per-minute LLM cost of the model you pick, plus your carrier; measure it in the pilot rather than trusting the plan price. Enterprise is custom-priced and is where BAAs for HIPAA, custom DPAs and SLAs, data residency in the US, EU, or India, and higher concurrency live.
Best for
CX and support-engineering leads at mid-market and enterprise companies who have engineers to own the agent and whose callers span many languages — fintech, marketplaces, telecom, and consumer subscription businesses running first-line voice support for informational and account-status calls. It is the strongest pick in the developer layer when voice quality and language coverage decide the project.
Skip it if nobody on your team will own prompts, tool calls, and evaluation, if you need a vendor to deliver the deployment end to end, or if most of your volume is chat — an ElevenLabs agent can type, but the reason to buy it is the voice.
Versus the alternatives
The two defaults in developer voice platforms are Retell AI and Vapi. Pick Retell when you want a bundled per-minute quote ($0.07–$0.31) and a choice of voice vendors; pick Vapi when engineers want to assemble speech, LLM, and telephony separately and accept a harder forecast in exchange for a $0.05 orchestration fee. ElevenLabs is the fastest-growing of the three; pick it when the voice is the part customers will judge you on.
If you want a vendor to own delivery, buy managed instead: PolyAI for healthcare and hospitality systems of record, Parloa for EU financial services and BPO-run contact centers, Sierra when the agent must transact across chat and voice under outcome-based billing. Decagon builds its own voice channel on ElevenLabs, so pick it when chat deflection is the main job and voice is the second channel.
If none fit, keep humans on the phone and put an ElevenLabs agent on after-hours overflow for your top three informational intents only.
Watch-outs
- HIPAA and zero-retention workspaces cannot use MCP servers. ElevenLabs’ own docs say this holds regardless of the agent’s MCP setting. Guard: if PHI or zero retention is in scope, design every tool call as a direct webhook or server tool from day one, and confirm the restriction’s current status in writing before architecture sign-off.
- The plan price hides the LLM and burst lines. Guard: run a two-week pilot on your real intent mix, record LLM cost per minute by model, size peak concurrency from your ACD data so burst stays the exception, and price Enterprise against that measured all-in rate.
- The headline results are vendor-published and scoped to easy calls. Klarna’s 10x and Revolut’s 8x both come from ElevenLabs case studies, and both deployments route complex calls to humans. Guard: set a containment target per intent against your human baseline, and judge the pilot on containment and transfer rate, not on speed.
- Paid credits expire on downgrade or cancellation. Guard: stay on monthly billing until pilot volume is stable, then move to an annual or Enterprise commitment.