Voice Agent Plans
Run a voice agent on a flat per-minute plan — ₹4, ₹5 or ₹6 a minute (₹8 for Best latency from 30 November 2026) for listening, the language model and the voice — instead of choosing and paying for each model yourself.
Overview
A voice agent plan (a tier) is a priced product: you choose the plan and a voice, and the platform chooses the speech recognition, the language model and the voice model. The whole AI layer of the call bills at one flat rate a minute.
The alternative is a custom stack: you set stt_model, voice_model and tts_model yourself, and each one bills at its own published rate, by the second, plus a flat ₹1 a minute platform fee on the call (see Custom stack billing).
| Plan | voice_tier | Per minute | Listening options | Best for |
|---|---|---|---|---|
| Standard | t1 | ₹4 (4 credits) | multilingual | Natural Indian-language voices at the lowest price |
| Expressive | t2 | ₹5 (5 credits) | multilingual | Richer, more expressive voices across Indian and world languages |
| Best latency | t3 | ₹6 (6 credits); ₹8 (8 credits) for calls that start on or after 30 November 2026, 00:00 IST | multilingual, english | The fastest replies and the best-sounding voices |
Standard speaks with Gnani Timbre v3, with 12 voices that each speak every supported language. See Voices.
1 credit = ₹1 ≈ US$0.0104 (US$1 = ₹96). Always read the live list from GET /api/v1/bots/config-schema (the voice_tiers array) rather than hard-coding it. Each tier's price_credits_per_minute is the price a call starting now pays; scheduled_price_change ({ "effective_at", "price_credits_per_minute" }, or null) announces a price change that has not started yet.
How a plan call is billed
- One flat rate a minute for the call's duration, charged once when the call ends.
- 30-second minimum per call. A 10-second call bills 30 seconds; a 2-minute call on Standard bills ₹8.
- A call that never connects costs nothing. A zero-length call does not reach the minimum.
- No per-component charges. A plan call is not also billed per speech-recognition second, per token or per character.
- No platform fee. The ₹1 a minute platform fee applies only to calls on a custom stack, never to a plan call.
- Phone carriage is separate. A rented number, or Meta's fee on a business-initiated WhatsApp call, is billed the same way as on a custom stack.
- The call ends when your credits run out. The running cost is checked against your balance during the call, and the call is stopped once your balance can no longer cover the minutes already spoken.
Put an agent on a plan
Set voice_tier (and optionally voice and voice_tier_speech_input) with a config patch. Requires bots:write.
curl -X PATCH https://api.callmissed.com/api/v1/bots/b1f2c3d4-5678-90ab-cdef-1234567890ab/config \
-H "Authorization: Bearer cm_your_api_key" \
-H "Content-Type: application/json" \
-d '{ "values": { "voice_tier": "t3", "voice_tier_speech_input": "english", "language": "en" } }'| Key | Values | If omitted |
|---|---|---|
voice_tier | t1, t2, t3 | The agent runs its own models and bills per component, plus the ₹1 a minute platform fee |
voice_tier_speech_input | A listening option the plan offers (see the table above) | The plan's first listening option |
voice | One of the plan's voices, listed per plan in voice_tiers[].voices | The plan's voice model uses its own default speaker |
language | multi (Multilingual, auto-detect) or one of the plan's languages in voice_tiers[].languages | The stack's own default (English) |
On a plan, the agent's own stt_model, voice_model, tts_model and tts_provider are ignored, not merged — a half-overridden plan would no longer match its price. voice and language stay yours. POST /api/v1/bots/validate-config reports a finding when a config sets both.
To move an agent back to a custom stack, remove the key:
curl -X PATCH https://api.callmissed.com/api/v1/bots/b1f2c3d4-5678-90ab-cdef-1234567890ab/config \
-H "Authorization: Bearer cm_your_api_key" \
-H "Content-Type: application/json" \
-d '{ "unset": ["voice_tier", "voice_tier_speech_input"] }'Custom stack billing
A call that is not on a plan is metered: each component bills at its published rate (speech recognition by the second, the language model per token, the voice per character), and the call adds one more line, the platform fee:
- ₹1 a minute (1 credit), charged once when the call ends, per second above the same 30-second minimum a plan call uses.
- It applies to every metered voice call: web and API voice sessions, the managed voice agent, and phone and WhatsApp calls whose agent bills per component.
- It appears as its own
voice_platform_feeline in the call's cost breakdown and in your usage history. - Component rates are the published list prices and do not change. Phone and WhatsApp network carriage is billed separately, as before.
Your per-minute total is the sum of the component lines for the models you chose, plus the ₹1 platform fee.
Languages
Standard and Expressive listen in 22 Indian languages plus English, including code-mixed speech such as Hinglish.
Best latency hears ten languages: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch. For any other language, use Standard or Expressive.
Multilingual (auto-detect)
Set "language": "multi" to make an agent multilingual on any plan. The call opens in English. After each thing the caller says, the agent detects the language and replies in the language the caller last spoke, switching mid-call if the caller switches. It can switch to any language in the plan's languages list. If the caller speaks a language the plan cannot speak, the agent keeps replying in the current language.
curl -X PATCH https://api.callmissed.com/api/v1/bots/b1f2c3d4-5678-90ab-cdef-1234567890ab/config \
-H "Authorization: Bearer cm_your_api_key" \
-H "Content-Type: application/json" \
-d '{ "values": { "voice_tier": "t1", "language": "multi" } }'On Best latency, multilingual needs the multilingual listening option. The english option hears English only, so the agent stays in English.
Custom (non-plan) agents accept "language": "multi" too. POST /api/v1/bots/validate-config warns about any chosen speech-recognition or voice model that cannot follow the caller's language. GET /api/v1/bots/config-schema lists the models that can, under multilingual.stt and multilingual.tts.
Read the plans
GET /api/v1/bots/config-schema returns the plans beside the per-model keys. Requires bots:read.
{
"voice_tiers": [
{
"id": "t1",
"label": "Standard",
"blurb": "Natural Indian-language voices at the lowest price.",
"price_credits_per_minute": 4.0,
"speech_inputs": [{ "id": "multilingual", "label": "Multilingual" }],
"voices": ["…"],
"voices_searchable": false,
"languages": ["en-IN", "hi-IN", "…"],
"multilingual": true,
"billing": "flat",
"minimum_seconds": 30.0
}
]
}| Field | Meaning |
|---|---|
id | The value to set as voice_tier |
price_credits_per_minute | The flat rate. 1 credit = ₹1 |
speech_inputs | Listening options; the first is the default |
voices | The featured voices this plan offers |
voices_searchable | true: the plan also accepts any voice from its voice model's full library (GET /api/v1/models/sonic-3.6/voices), not only the featured ones |
languages | The languages this plan can hear and speak, English first. language takes one of these or multi |
multilingual | true: the plan accepts "language": "multi" |
billing | Always flat |
minimum_seconds | The billed floor per connected call |
Where to see what a call cost
GET /v1/voice/sessions/{id}/cost returns the cost of a finished session, and your usage history lists plan calls as one line per call.