Version: 1.0.x

Sarvam AI LLM

The Sarvam AI LLM provider enables your agent to use Sarvam AI's language models for text-based conversations and processing.

Installation

Install the Sarvam AI-enabled VideoSDK Agents package:

pip install "videosdk-plugins-sarvamai"

Importing

from videosdk.plugins.sarvamai import SarvamAILLM

note

When using Sarvam AI as the LLM option, the function tool calls and MCP tool will not work.

Authentication

The Sarvam plugin requires a Sarvam API key.

Set SARVAM_API_KEY in your .env file.

Example Usage

from videosdk.plugins.sarvamai import SarvamAILLM
from videosdk.agents import Pipeline

# Initialize the Sarvam AI LLM model
llm = SarvamAILLM(
    model="sarvam-105b",
    # When SARVAMAI_API_KEY is set in .env - DON'T pass api_key parameter
    api_key="your-sarvam-ai-api-key",
    temperature=0.7,
    tool_choice="auto",
    max_completion_tokens=1000,
    reasoning_effort="medium",  # Optional: "low", "medium", "high"
    wiki_grounding=False,       # Optional: enable Wikipedia-grounded responses
    top_p=1,
    frequency_penalty=0,
    presence_penalty=0,
)

#  Add llm to pipeline
pipeline = Pipeline(llm=llm)

note

When using .env file for credentials, don't pass them as arguments to model instances or context objects. The SDK automatically reads environment variables, so omit api_key and other credential parameters from your code.

Configuration Options

model: (str) The Sarvam AI model to use (default: "sarvam-m").
api_key: (str) Your Sarvam AI API key. Can also be set via the SARVAMAI_API_KEY environment variable.
temperature: (float) Sampling temperature for response randomness (default: 0.7).
tool_choice: (ToolChoice) Tool selection mode (default: "auto").
max_completion_tokens: (int) Maximum number of tokens in the completion response (optional).
reasoning_effort: (str) Controls reasoning depth for the model. Allowed values: "low", "medium", "high" (default: None).
wiki_grounding: (bool) Enables Wikipedia search to ground responses with factual information (default: False).
top_p : An alternative to sampling with temperature. Defaults to None.
frequency_penalty: Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far.
presence_penalty: Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far
stop : Up to 4 sequences where the API will stop generating further tokens. The returned text will not contain the stop sequence.