Three ways to work with us. All figures indicative.

Vapi and Retell are orchestration layers. You pay them, and then you separately pay for speech recognition, a language model, a voice provider, and telephony. Their headline per-minute number is real, but it is not your bill. Ours is a single rate because we own the stack end to end, so ASR, the model, the voice, and the telephony are already inside the $0.09.* The honest caveat: if you have already negotiated hard with each of your underlying vendors, the gap narrows. Where it does not narrow is dialect. We are not claiming to out-sound them in English. We are claiming parity in English plus city-level registers they do not offer, plus mid-sentence language switching, at roughly a third of an assembled stack's cost. Bring your last invoice to a migration audit and we will do the comparison line by line rather than asking you to trust a table.

Yes, and we would rather set expectations honestly than dangle a tier list. Committed volume moves the per-minute rate, and multi-year commitments move it further. We have not locked the published breakpoints yet, so we are not going to invent thresholds on a public page just to look generous. What we can say now: the $0.09 figure* is the indicative starting point rather than a floor, discounts are agreed against committed minutes rather than forecast minutes, and the deployment fee is usually the more negotiable line for high-volume accounts. Bring your real annual minutes to the workshop and you will get a real number.

Our warranty and service credit terms are still with legal, so we are not going to describe them here and have you plan around wording that may change. What we can tell you today is the shape of it. A 99.99% SLA is available and is agreed per deployment rather than assumed. Enterprise agreements carry a defined support response commitment. Anything beyond that, including remedies and service credits, is confirmed in the contract and nowhere else.

CLAIM PENDING · warranty terms in legal review

Because a voice deployment that works is not a signup form. The fee covers telephony routing into your numbers, integration with your CRM or ticketing system, voice and dialect tuning against your own call recordings, adversarial testing before you put a customer on it, and a human on the other end during go-live week. Vendors who skip the fee usually push that work onto your team and call it self-serve. If your deployment is genuinely simple, say so in the workshop and we will scope it down rather than defend the floor.*

Carrier and number costs in some markets, depending on how your telephony is routed. Custom API work outside the agreed integration scope, which is quoted separately on Platform Resell and included on Custom Build. Any compliance certification work specific to your regulator rather than to our platform. We would rather list these now than surprise you at invoice three.

Yes. Most enterprises start with one language, one use case, and one queue, prove it against their own numbers, and expand from there. That is also the cheapest way to find out whether we are wrong about your market. Start narrow, measure it against the agents you already have, and only widen the scope once the comparison is boring.

Probably not. Agni runs self-serve in India at ravan.ai, with its own pricing and its own onboarding. RavanAI Enterprises is the international enterprise deployment arm. If you are an Indian enterprise with international operations, talk to us anyway and we will point you to whichever door costs you less.