langchain-sarvamai

The langchain-sarvamai package provides a LangChain integration for Sarvam AI's language models, optimised for Indian languages and enterprise workloads.

Installation

pip install langchain-sarvamai

Setup

Get your API key from Sarvam AI and set it as an environment variable:

export SARVAM_API_KEY="your-api-key"

Usage

Chat Model

from langchain_sarvamai import ChatSarvam

llm = ChatSarvam(model="sarvam-105b")

# Invoke
response = llm.invoke("What is the capital of India?")
print(response.content)

# With system message
from langchain_core.messages import HumanMessage, SystemMessage

messages = [
    SystemMessage(content="You are a helpful assistant."),
    HumanMessage(content="Tell me about Indian languages."),
]
response = llm.invoke(messages)
print(response.content)

# Stream
for chunk in llm.stream("Count from 1 to 5."):
    print(chunk.content, end="", flush=True)

# Async
import asyncio
response = asyncio.run(llm.ainvoke("Hello!"))

Available Models

Reasoning Mode

llm = ChatSarvam(model="sarvam-105b", reasoning_effort="high")
response = llm.invoke("Explain the Riemann hypothesis.")
# Access reasoning trace
print(response.additional_kwargs.get("reasoning_content"))

Supported values for reasoning_effort: "low", "medium", "high".

Parameters

Development

# Install with dev dependencies
pip install -e ".[test]"

# Run unit tests
pytest tests/unit_tests/ -v

# Run integration tests (requires SARVAM_API_KEY)
pytest tests/integration_tests/ -v

License

MIT