A Retell AI alternative for Hindi and Hinglish phone agents

How do Retell AI and Voice Infra differ?

Retell is a voice-agent platform that lets you choose the parts. Its pricing page breaks a minute into the conversation engine (Retell Voice Infra at $0.055), the voice (from $0.015 for Retell’s own voices up to $0.10 for some ElevenLabs models) and the LLM, from $0.0016 to $0.64 a minute depending on the model and whether you pick its fast tier. Telephony is a separate line. The calculator’s default agent lands at $0.11 a minute before the phone line.

Voice Infra sells the whole call at one rate. Every tier runs the same Mira stack, with Mira STT and Mira TTS as our own models, and on T5 and T3 the rate includes Mirai phone calling and call insights. You choose a tier for the price you want to commit to, since the stack does not change between them. Minutes are billed in 10-second blocks, rounded up with a one-block minimum, and unanswered, busy and failed calls cost nothing.

What does calling India cost on each?

Retell’s call-rate menu lists India through Twilio at $0.15 a minute and through Telnyx at $0.25 a minute, alongside a note that SIP trunking and custom telephony carry no Retell charge. Retell-managed numbers are US and Canada only, so a number in India comes through custom telephony from your own provider. Taking the calculator default, 1,000 minutes carried on the India (Twilio) rate is 1,000 × ($0.11 + $0.15) = $260.

On T5, the same 1,000 minutes is 1,000 × ₹5 = ₹5,000 with the phone line included. At volume, T1 brings the rate to ₹1 a minute: its ₹1,00,000 monthly minimum buys 1,00,000 minutes, with Mirai phone calling at ₹0.30 a minute if you want us to carry the line. T3 sits between, with a ₹50,000 minimum that buys 16,666 minutes at ₹3 (₹50,000 ÷ ₹3). Prices exclude GST. We do not convert dollars to rupees here; use your bank’s rate. The pricing guide walks through a monthly forecast with the 10-second rounding.

What does moving a Retell agent look like?

The prompt, the opening line and the variables move first. On our API they are system_prompt, first_message and {{variables}} on POST /v2/agents. If your Retell agent is a flow, rebuild it as a node-based agent; Agent Studio can draft one from a description or from recordings of your calls, and the prompt versus node guide helps decide which shape fits. Calls go out with POST /v2/calls, campaigns dial an uploaded contact list with retries and a calling window, and results arrive through signed webhooks, which the webhook guide covers.

Plan around what we do not have yet. There is no transfer to a human today, and a new workspace starts at 5 concurrent calls until we raise it. If your Retell agents depend on warm transfer or IVR navigation, test that part of the journey before you commit.

What should you test before switching?

Start from real calls. Pull a handful of recordings of your own Hindi and Hinglish conversations (with consent) and turn them into test scripts that keep the hard parts: customer names, rupee amounts, dates, addresses and the English product terms people drop mid-sentence. Run the same script on both platforms against your own phone.

Then measure what a caller experiences. Time the gap from the caller’s last word to the first audio of the reply on a real phone line, and repeat it enough times to see the slow end as well as the typical one. The latency guide explains how to separate network, recognition, model and voice in that number. Finally, compare the bill: count billed minutes after rounding, add the phone line on both sides, and check that the outcome fields your team needs arrive with each call.