AI CUSTOMER AGENTS, BUILT AROUND YOUR BUSINESS
MULTILINGUAL CUSTOMER EXPERIENCE / IN DEVELOPMENT

English in.
A familiar answer out.

Let a customer ask in English and see how a business-grounded reply could read in Hindi, Marathi, or Tamil—with the original source kept in view.

Concept preview / curated translations / Sarvam API not connectedTry the language preview
नमस्तेनमस्कारவணக்கம்ONE KNOWLEDGE BASE / MANY WAYS TO BE HEARD
ONE ENGLISH QUESTION / A LOCAL-LANGUAGE REPLY

Hear the difference
in the answer.

Choose a response language. This is a curated example to demonstrate the experience; it does not call Sarvam or any AI model.

RESPONSE LANGUAGE
QUESTION / ENGLISH

Can I change my delivery address?

SAMPLE SOURCEDelivery policy / example only
CHATBHIDU / PRODUCT CONCEPTCURATED SAMPLE
b.
AGENT RESPONSE / HINDI

आप ऑर्डर भेजे जाने से पहले अपना डिलीवरी पता बदल सकते हैं। अगर आपका ऑर्डर पहले ही भेजा जा चुका है, तो कृपया सहायता टीम से संपर्क करें।

Delivery policy / sample source

Translations above are curated examples, not generated by Sarvam. Have native speakers validate product terminology and tone before production use.

TEXT FIRST / VOICE NEXTSEPARATE MODEL CAPABILITIES ↘
A PRACTICAL ROLLOUT

Start with translated text.
Add speech deliberately.

Text translation can be added to an existing retrieval flow. Voice is a separate product phase with audio capture, transcription, speech generation, consent, and more operational cost.

01 / TEXT CHAT

Answer in the customer's chosen language

Translate the final grounded answer; keep citations and source IDs separate and unchanged. Fall back clearly if translation fails.

FEASIBLE / FIRST
02 / SPEECH INPUT

Understand spoken questions

Sarvam Saaras v4 documents speech recognition for 22 Indian languages plus English. Browser microphone permission and audio handling still need design.

PLANNED
03 / SPOKEN REPLIES

Let the agent speak back

Sarvam Bulbul v3 documents speech generation for 10 Indian languages plus English. Coverage is narrower than speech recognition and must be checked per language.

PLANNED

Sarvam model coverage differs by capability. The three-language selector above is an illustrative product sample, not a claim of live provider support.

EARLY ACCESS / LANGUAGE RESEARCH

Help us build for
how people speak.

Tell us which language, workflow, and terminology matter to your customers. Native-speaker review will be part of validating this experience.

Share a language use case