sarvam.ai

•

API Key

About Sarvam AI

Sarvam AI is an Indian AI company providing a full-stack sovereign AI platform tailored to Indian languages. Its API suite covers speech-to-text, text-to-speech, translation, and chat completion via its Saaras LLM, all supporting 22+ Indian languages including Hindi, Bengali, Tamil, and Telugu. Sarvam handles code-switching (Hinglish, Tanglish) and accented speech that general-purpose global models struggle with. As of June 2026, Sarvam reached a $1.5B valuation after a $150M investment from HCLTech.

AI Agent Use Cases

- Speech-to-text and text-to-speech APIs for 22+ Indian languages

- Code-switching support for mixed-language speech (Hinglish, Tanglish)

- Sarvam-30B and Sarvam-105B foundational LLMs for Indian-language reasoning

- Translation API across all supported Indian languages

- Chat completion via the Saaras LLM, free at the API level

- On-premise and sovereign deployment options for regulated enterprises

Available Actions

These are the specific actions that AI agents can perform with this tool

Chat Completions

13 inputs

Get a chat completion from Sarvam LLM based on a list of messages and specified model.

Inputs

messages

A list of messages comprising the conversation so far.

model

Model ID used to generate the response. Only 'sarvam-m' is supported.

temperature

Sampling temperature. Higher values make output more random.

top_p

Nucleus sampling probability mass.

reasoning_effort

Controls the model's reasoning effort.

max_tokens

Maximum number of tokens generated in the chat completion.

stream

Whether to stream the response data as it's generated.

stop

Sequences where API will stop generating further tokens.

How many chat completion choices to generate for each input message.

seed

A seed for deterministic sampling.

frequency_penalty

Positive values penalize new tokens based on their frequency.

presence_penalty

Positive values penalize new tokens based on whether they appear in the text so far.

wiki_grounding

If enabled, model uses RAG from Wikipedia for better factual answers.

Real Time Speech To Text

3 inputs

Transcribe a given audio file (WAV or MP3) into text using real-time speech-to-text in Indian languages and English.

Inputs

file

The audio file to transcribe. Supported formats: WAV (.wav), MP3 (.mp3).

model

Specifies the speech-to-text model to use.

language_code

Language code of the input audio. If not provided, the API will auto-detect.

Realtime Speech To English Translator

3 inputs

Transcribe an audio file and translate its speech to English, optionally supplying a prompt and choosing a specific model.

Inputs

file

The audio file containing speech to transcribe and translate to English

prompt

Conversation context prompt to boost model accuracy

model

Model to use for converting speech to text in the target language

Text To Speech Conversion

9 inputs

Convert input text into spoken audio using text to speech

Inputs

text

The input text to be converted to speech, must be less than or equal to 1500 characters.

target_language_code

The BCP-47 language code representing the language of the text.

speaker

The speaker's voice to use in output audio. Defaults to Anushka. Speaker must be compatible with model.

pitch

Adjusts the pitch of the output audio. Lower is deeper, higher is sharper. Default is 0.0.

pace

Controls speech speed. Lower is slower, higher is faster. Default is 1.0.

loudness

Controls audio loudness. Lower is quieter, higher is louder. Default is 1.0.

speech_sample_rate

The sample rate of output WAV audio. Default is 22050 Hz.

enable_preprocessing

Whether to enable normalization (for mixed-language text and numbers/dates). Default is false.

model

Specifies the model to use for TTS conversion. Defaults to bulbul:v2.

Translate Text Between Languages

9 inputs

Translate input text from one language to another using Indian language codes, with options for translation style, model, script, and formatting, returning the translated text and related metadata.

Inputs

input

The input text to translate. Must not exceed 2000 characters.

source_language_code

Source language code of the input text.

target_language_code

Language code for the translation target.

speaker_gender

Optionally specify the speaker's gender for improved translation.

mode

The tone or style of translation. Default is 'formal'.

model

Optional translation engine. Choose 'mayura:v1' or 'sarvam-translate:v1'.

enable_preprocessing

When true, enables preprocessing for better translation of complex text.

output_script

Transliteration style for output text (mayura:v1 only).

numerals_format

Preferred numerals format for output (international or native).

About Sarvam AI

AI Agent Use Cases

- Speech-to-text and text-to-speech APIs for 22+ Indian languages

- Code-switching support for mixed-language speech (Hinglish, Tanglish)

- Sarvam-30B and Sarvam-105B foundational LLMs for Indian-language reasoning

- Translation API across all supported Indian languages

- Chat completion via the Saaras LLM, free at the API level

- On-premise and sovereign deployment options for regulated enterprises

Available Actions

These are the specific actions that AI agents can perform with this tool

Chat Completions

13 inputs

Get a chat completion from Sarvam LLM based on a list of messages and specified model.

Inputs

messages

A list of messages comprising the conversation so far.

model

Model ID used to generate the response. Only 'sarvam-m' is supported.

temperature

Sampling temperature. Higher values make output more random.

top_p

Nucleus sampling probability mass.

reasoning_effort

Controls the model's reasoning effort.

max_tokens

Maximum number of tokens generated in the chat completion.

stream

Whether to stream the response data as it's generated.

stop

Sequences where API will stop generating further tokens.

How many chat completion choices to generate for each input message.

seed

A seed for deterministic sampling.

frequency_penalty

Positive values penalize new tokens based on their frequency.

presence_penalty

Positive values penalize new tokens based on whether they appear in the text so far.

wiki_grounding

If enabled, model uses RAG from Wikipedia for better factual answers.

Real Time Speech To Text

3 inputs

Transcribe a given audio file (WAV or MP3) into text using real-time speech-to-text in Indian languages and English.

Inputs

file

The audio file to transcribe. Supported formats: WAV (.wav), MP3 (.mp3).

model

Specifies the speech-to-text model to use.

language_code

Language code of the input audio. If not provided, the API will auto-detect.

Realtime Speech To English Translator

3 inputs

Transcribe an audio file and translate its speech to English, optionally supplying a prompt and choosing a specific model.

Inputs

file

The audio file containing speech to transcribe and translate to English

prompt

Conversation context prompt to boost model accuracy

model

Model to use for converting speech to text in the target language

Text To Speech Conversion

9 inputs

Convert input text into spoken audio using text to speech

Inputs

text

The input text to be converted to speech, must be less than or equal to 1500 characters.

target_language_code

The BCP-47 language code representing the language of the text.

speaker

The speaker's voice to use in output audio. Defaults to Anushka. Speaker must be compatible with model.

pitch

Adjusts the pitch of the output audio. Lower is deeper, higher is sharper. Default is 0.0.

pace

Controls speech speed. Lower is slower, higher is faster. Default is 1.0.

loudness

Controls audio loudness. Lower is quieter, higher is louder. Default is 1.0.

speech_sample_rate

The sample rate of output WAV audio. Default is 22050 Hz.

enable_preprocessing

Whether to enable normalization (for mixed-language text and numbers/dates). Default is false.

model

Specifies the model to use for TTS conversion. Defaults to bulbul:v2.

Translate Text Between Languages

9 inputs

Translate input text from one language to another using Indian language codes, with options for translation style, model, script, and formatting, returning the translated text and related metadata.

Inputs

input

The input text to translate. Must not exceed 2000 characters.

source_language_code

Source language code of the input text.

target_language_code

Language code for the translation target.

speaker_gender

Optionally specify the speaker's gender for improved translation.

mode

The tone or style of translation. Default is 'formal'.

model

Optional translation engine. Choose 'mayura:v1' or 'sarvam-translate:v1'.

enable_preprocessing

When true, enables preprocessing for better translation of complex text.

output_script

Transliteration style for output text (mayura:v1 only).

numerals_format

Preferred numerals format for output (international or native).