Comparison
A comparison with the vendor we run on ourselves.
The voice and the transcription in the one real call quoted across this site are ElevenLabs. Their agents platform does, over the same models, what we do — and this page compares two products, one of which runs on the other.
Choose ElevenLabs Agents if you want the best voice catalogue on the market, a minute that already includes speech and transcription, and a vendor with large customers. Choose CallAgent if you need human operators and a switch in the same product, an install on your own server, or the freedom to change speech vendor without changing platform.
The rule this page is written under: every row has a sentence in both columns; everything we say about ElevenLabs is read off elevenlabs.io and their documentation; and the fact that we are their customer is in the first line of the page rather than in a footnote. Six rows out of eleven go to them.
The voice vendor, and the product built on top of it
CallAgent is for you if…
You need your own people to take the call over from the agent, in queues, with dispositions and a supervisor. Or you want the product installed on your own infrastructure. Or you want to be able to change speech vendor later without rewriting everything you built.
ElevenLabs Agents is for you if…
Voice quality and voice choice are the main criterion — they are the vendor we use ourselves. Or you want a minute with speech and transcription already included, and a free tier of fifteen minutes to start from today.
Eleven dimensions, six of them lost
We lose the voices row by definition: our speech vendor is them. It stays in the table because that is precisely what makes the comparison worth making.
| Dimension | CallAgent | ElevenLabs Agents |
|---|---|---|
| what it actually is | A voice agent platform, a SIP PBX and a contact centre, in one console and on one invoice. The voice and the transcription are bought from them. | An agent platform built on their own speech and transcription models, with a visual builder, SDKs and a web widget. |
| voices ElevenLabs Agents wins this row | Exactly one speech vendor open today: them. The rest have adapter code and are marked not implemented. | They advertise over five thousand voices on the agents platform. It is their core product and there is no way for us to win this row. |
| human operators | Queues, skills, a browser softphone, dispositions, wrap-up, and a human taking the call over without the caller being rung again. A supervisor console with listen, whisper and barge. | Transfer to a number or a SIP endpoint, in three modes: a conference with a whisper message, a blind transfer and SIP REFER. Destinations have to be pre-configured in the agent's transfer rules; queues and skills-based routing do not appear in their documentation. |
| telephony ElevenLabs Agents wins this row | Twilio works end to end, with numbers bought or imported from your own account. Our own SIP trunks are built, but an inbound call over a trunk does not reach an assistant yet. | A documented bring-your-own SIP trunk, inbound and outbound, with digest authentication, TLS and SRTP. They sell no numbers at all: you bring them from Twilio or from your own trunk. |
| how the price is built ElevenLabs Agents wins this row | €0.10 per minute on top of provider costs at their price — theirs included — or €0.10 and nothing else if you bring your own keys. | Subscriptions with included minutes, from 15 free minutes up to 12,375 minutes for $990 a month, and $0.08 a minute over them. Speech and transcription are included in the minute; the language model and the telephony are billed separately. |
| concurrent calls ElevenLabs Agents wins this row | We publish no limit, because we have never run concurrent volume. That does not mean "unlimited", it means "unmeasured". | The limit is written on every plan, from four concurrent calls on the free tier to forty on the $990 one. Above it you can burst to triple, at double the rate and with reduced processing priority. |
| where it runs | Hosted, or installed on your own server in containers with a documented install. Recordings and transcripts can be written straight into your own bucket, across eight destinations. | Their hosted service only. EU data residency exists, but their documentation describes it as an Enterprise feature, and restricting processing to the EU additionally requires Zero Retention Mode and using the API. |
| Romanian | The console, the guides, the billing and the documentation are in Romanian, and the personal-data masking recognises the Romanian CNP. The language is set per assistant and has to be set explicitly. | Romanian is named in the language list of the very speech model we use, and their transcription places it in the top accuracy tier. Their agents language page does not enumerate its languages, and the product is in English. |
| certifications ElevenLabs Agents wins this row | None. There is no external audit and no third-party penetration test has been done. | SOC 2 and GDPR stated on their enterprise page, plus a HIPAA business associate agreement for eligible customers and a zero retention mode that stores no recordings or transcripts. On ISO 27001 and PCI we could read nothing: their compliance centre does not open without JavaScript. |
| real traffic ElevenLabs Agents wins this row | Eleven real calls, six of which failed, and no customer served. We have run no volume and taken no real payment. | Case studies with Cisco, NVIDIA, TELUS Digital and Bertelsmann on their enterprise page, and offices in four cities. They do not publish a total call volume. |
| cost transparency | Every call has per-stage cost rows stored immutably: quantity, unit cost, where the quantity came from, and whether the price was verified against the vendor. | Their post-call webhook sends a single cost number, with no split between model, speech, transcription and telephony. |
The ElevenLabs figures are read from elevenlabs.io/pricing/agents, elevenlabs.io/enterprise and their documentation in September 2026. We assert nothing about their margin over raw speech pricing: their two price lists are published in different units and are never reconciled.[1]
When not to pick us
Four situations where the right answer is to go straight to them — and one of them is our own architecture.
What we do differently
Four things, each checkable in the product or in the code.
A contact centre included
Queues, operators and the supervisor are in the same product as the agent, not a second vendor to wire in.
Runs on your machine
A documented container install. With them, EU residency is an Enterprise feature and a customer-side install does not exist.
A Romanian product
Console, guides, billing and CNP masking. Not a translation of an English product.
Cost per stage
Immutable rows on every call, instead of a single cost number in a webhook.
What people ask when comparing the two
If you use ElevenLabs, why not go straight to them?
If the voice agent is all you need, that is exactly what you should do, and it says so in the list above. Our layer only pays for itself if you need what they do not have: queues, human operators with a softphone, dispositions, a supervisor with whisper, a SIP switch, an install on your own server, and the cost itemised on every call.
Which one works out cheaper?
It depends on volume and does not compare directly. With them you pay a subscription with included minutes — from free, with fifteen minutes, up to $990 for 12,375 — and $0.08 a minute above it; speech and transcription are in the minute, the model and the telephony are billed separately. With us the platform takes €0.10 a minute, and speech, transcription, model and telephony go in at the vendor's price. We do not say what they add over raw speech pricing, because they do not publish the two prices in the same unit.
Can I change speech vendor later?
With us that is the idea, although today it is more promise than fact: the model, the transcriber and the voice are chosen per assistant, and the registry holds adapter code for several vendors — but exactly one is open today, and it is them. With them, speech and transcription are the product, so the question does not arise.
Where do the recordings sit?
With us, in a bucket of your own if you want, on any of the eight supported destinations, or on your own server if you install the product there. With them, EU residency is described in the documentation as an Enterprise feature, and restricting processing to the EU additionally requires Zero Retention Mode and use of the API.
Is it not odd to compare yourself with your own supplier?
It is, which is why it is in the first sentence of the page. A reader comparing us with them would find out within five minutes that the voice in our one real call is theirs. We would rather it were the first thing they read than something they discover on their own and then use to doubt the rest of the page.
What to do next
If all you need is the agent, go to them. If you need what comes after the call as well, read the page that says what does not work here first.
What else people compare us with
If you got this far you probably have two other tabs open. They are written under the same rule: sentences in both columns, and the rows we lose left in the table.
- CallAgent vs Vapi For somebody who wants only the platform, as cheap per minute as possible, with nothing to administer.
- CallAgent vs Retell AI For somebody building the agent quickly and connecting it to the contact centre they already run.
- CallAgent vs Bland AI For somebody who wants one per-minute price with the model and the voice included, and channels beyond the telephone.
- Synthflow alternative For somebody with an enterprise budget, a security questionnaire to pass, and a preference for a European vendor.
- Genesys alternative For somebody buying an enterprise contact centre, with everything in it and an implementation project around it.
If you need more than the voice
You already have the voice vendor on their own page, with a free tier. We are worth a conversation only if you need queues, operators, a switch or an install of your own. Tell us which.
Notes
- 1 The ElevenLabs Agents facts: their public pages elevenlabs.io/pricing/agents, elevenlabs.io/enterprise and their documentation — models, speech to text, SIP trunking, transfer, data residency, zero retention mode and the post-call webhook, read in September 2026.
- 2 The real call quoted across this site used ElevenLabs scribe_v2_realtime for transcription and eleven_flash_v2_5 for the voice, with Anthropic claude-sonnet-4-6 for the model. The call identifier and its events are on the status page.