sarvam-ai-sdk weekly downloads statistics

<a href="https://github.com/sarvamai/sarvam-ai-sdk"> <img alt="cover" src="https://github.com/sarvamai/sarvam-ai-sdk/blob/master/cover.png?raw=true" /> </a>

AI SDK - Sarvam Provider

The Sarvam provider for the AI SDK contains language model support for the Sarvam chat completion, Text-to-Speech and Speech-to-Text APIs.

Setup

The Sarvam provider is available in the sarvam-ai-sdk module. You can install it with

npm i sarvam-ai-sdk ai@7

[!WARNING] This package only works with Vercel AI-SDK v7, not v8. Make sure to install ai@7 in your project.

Version Compatibility

Sarvam AI SDK Version	Vercel AI SDK Version
0.5.x (beta)	8.x.x (beta)
0.4.x (current)	7.x.x (current)
0.3.x	6.x.x
0.2.x	5.x.x
0.1.x	4.x.x

Provider Instance

You can import the default provider instance sarvam from sarvam-ai-sdk:

import { sarvam } from "sarvam-ai-sdk";

Create .env file with API key from Sarvam Dashboard

SARVAM_API_KEY="your_api_key"

Example

import { sarvam } from "sarvam-ai-sdk";
import { generateText } from "ai";

const { text } = await generateText({
	model: sarvam("sarvam-105b", {
		reasoning_effort: "low",
	}),
	prompt: "Translate this to malayalam: 'Keep cooking, guys'",
});

console.log(text); // പാചകം തുടരൂ, സുഹൃത്തുക്കളേ

Text-to-Speech

import { sarvam } from "sarvam-ai-sdk";
import { generateSpeech } from "ai";
import { writeFile } from "fs/promises";

const { audio } = await generateSpeech({
	model: sarvam.speech("bulbul:v3", "ml-IN"),
	text: "പാചകം തുടരൂ, സുഹൃത്തുക്കളേ",
});

const audioBuffer = Buffer.from(audio.base64, "base64");
await writeFile("./src/transcript-test.wav", audioBuffer);

Speech-to-Text

import { sarvam } from "sarvam-ai-sdk";
import { transcribe } from "ai";
import { readFile } from "fs/promises";

const { text } = await transcribe({
	model: sarvam.transcription("saaras:v4", "en-IN"),
	audio: await readFile("./src/transcript-test.wav"),
});

console.log(text); // Pachakam thudaroo, suhruthukkale.

Translation

NB: Only translates prompt and role:user messages, not role:system not role:assistant.

import { sarvam } from "sarvam-ai-sdk";
import { generateText } from "ai";

const result = await generateText({
	model: sarvam.translation("mayura:v1", {
		from: "ml-IN",
		to: "en-IN",
	}),
	prompt: "ഇതൊക്കെ ശ്രദ്ധിക്കണ്ടേ അംബാനെ?",
});

console.log(result.text); // Shouldn't we be careful about this, Ambane?

Transliterate

NB: Only transliterates prompt and role:user messages, not role:system not role:assistant.

import { sarvam } from "sarvam-ai-sdk";
import { generateText } from "ai";

const result = await generateText({
	model: sarvam.transliterate({
		to: "ml-IN",
		from: "en-IN", // optional
	}),
	prompt: "eda mone, happy alle?",
});

console.log(result.text); // എടാ മോനെ, ഹാപ്പി അല്ലേ?

Language Identification

NB: Only identifies prompt and role:user messages, not role:system not role:assistant.

import { sarvam } from "sarvam-ai-sdk";
import { generateText } from "ai";

const result = await generateText({
	model: sarvam.languageIdentification(),
	prompt: "ബുദ്ധിയാണ് സാറേ ഇവൻ്റെ മെയിൻ",
});

console.log(result.text); // ml-IN

Tool Calling

import { z } from "zod";
import { generateText, tool } from "ai";
import { sarvam } from "sarvam-ai-sdk";

const result = await generateText({
	model: sarvam("sarvam-105b"),
	tools: {
		weather: tool({
			description: "Get the weather in a location",
			inputSchema: z.object({
				location: z.string(),
			}),
			execute: async ({ location }) => ({
				location,
				temperature: 72 + Math.floor(Math.random() * 21) - 10,
			}),
		}),
	},
	instructions: "Your are a helpful AI",
	prompt: "കൊച്ചിയിലെ കാലാവസ്ഥ എന്താണ്?",
});

console.log(result.toolResults);

Generating Structured Outputs

import { z } from "zod";
import { sarvam } from "sarvam-ai-sdk";
import { generateText, Output } from "ai";

const { output } = await generateText({
	model: sarvam("sarvam-105b", {
		experimental_json_mode: true, // to enable direct JSON mode
		reasoning_effort: "none", // to reduce inference time
	}),
	output: Output.object({
		name: "Recipe",
		description: "A recipe with a name, ingredients and steps",
		schema: z.object({
			recipe: z.object({
				name: z.string(),
				ingredients: z.array(z.string()),
				steps: z.array(z.string()),
			}),
		}),
	}),
	prompt: "Generate a South Indian recipe, in Malayalam",
});

console.log(output);

EVE Agents

Build durable agents with Vercel's Eve Framework.

import { simulateStreamingMiddleware, wrapLanguageModel } from "ai";
import { defineAgent } from "eve";
import { sarvam } from "sarvam-ai-sdk";
import { SarvamChatModelInfo } from "sarvam-ai-sdk/info";

const model = wrapLanguageModel({
	model: sarvam("sarvam-105b"),
	middleware: simulateStreamingMiddleware(),
});

export default defineAgent({
	model,
	modelContextWindowTokens: SarvamChatModelInfo["sarvam-105b"].context_window,
});

All APIs

import { sarvam, createSarvam } from "sarvam-ai-sdk";

// Override `SARVAM_API_KEY` using
createSarvam({ apiKey: "your_api_key" })

// Text-to-Text + Chat Completion
sarvam("sarvam-105b-conversations");
sarvam.languageModel("sarvam-105b", {
	version: "v2",
});

// Text-to-Text + Transliteration
sarvam.transliterate({ to: "ml-IN", from: "en-IN" });

// Text-to-Text + Translation
sarvam.translation("mayura:v1", { from: "en-IN", to: "ml-IN" });

// Text-to-Text + Language identification
sarvam.languageIdentification();…