Skip to main content

Anthropic AI Provider (@voltagent/anthropic-ai)

The Anthropic AI Provider integrates VoltAgent with Anthropic's Claude models. It wraps the official @anthropic-ai/sdk SDK.

Key Characteristics:

  • Claude Models: Supports all Claude models through the Anthropic API.
  • Model Agnostic: Accepts standard Claude model identifier strings (e.g., 'claude-3-sonnet-20240229', 'claude-3-opus-20240229').
  • Core Functionality: Focuses on text generation, streaming, and structured object generation using the underlying Anthropic SDK.

Installation​

npm install @voltagent/core @voltagent/anthropic-ai zod

Note: @anthropic-ai/sdk is a peer dependency. zod is required if using generateObject.

Configuration​

The AnthropicProvider requires an API key, which can be provided directly or through an environment variable.

import { AnthropicProvider } from "@voltagent/anthropic-ai";

// Option 1: Direct API Key
const anthropicProvider = new AnthropicProvider({
  apiKey: "YOUR_ANTHROPIC_API_KEY",
});

// Option 2: Environment Variable (ANTHROPIC_API_KEY)
// Ensure process.env.ANTHROPIC_API_KEY is set
const anthropicProvider = new AnthropicProvider({});

Usage​

Instantiate your Agent with the configured AnthropicProvider:

import { Agent } from "@voltagent/core";
import { AnthropicProvider } from "@voltagent/anthropic-ai";

const anthropicProvider = new AnthropicProvider({ apiKey: "YOUR_ANTHROPIC_API_KEY" });

const agent = new Agent({
  name: "Claude Agent",
  instructions: "An agent powered by Claude",
  llm: anthropicProvider,
  model: "claude-3-sonnet-20240229", // Specify the desired Claude model
});

// Example call
async function run() {
  const response = await agent.generateText("What is the capital of France?");
  console.log(response.text);
}

run();

Supported Methods​

  • generateText: ✅ Supported. Calls Anthropic SDK's messages.create.
  • streamText: ✅ Supported. Calls Anthropic SDK's messages.create with stream: true.
  • generateObject: ✅ Supported. Uses Anthropic's structured output capabilities with Zod schema validation.
  • streamObject: ✅ Supported. Streams structured objects with schema validation.

Tool Calling Support​

✅ Supported.

The @voltagent/anthropic-ai provider supports tool calling through the MCP (Model Control Protocol) format:

import { createTool } from "@voltagent/core";
import { z } from "zod";

const weatherTool = createTool({
  name: "get_current_weather",
  description: "Get the current weather in a location",
  parameters: z.object({
    location: z.string().describe("The location to get weather for"),
  }),
  execute: async (input) => {
    return {
      location: input.location,
      temperature: 72,
      condition: "sunny",
    };
  },
});

const agent = new Agent({
  name: "weather-agent",
  instructions: "A helpful weather assistant",
  llm: new AnthropicProvider(),
  model: "claude-3-sonnet-20240229",
  tools: [weatherTool],
});

Model Selection & Options​

The specific Claude model (e.g., 'claude-3-sonnet-20240229', 'claude-3-opus-20240229') is set via the model property during Agent instantiation.

You can override or provide additional Anthropic-specific generation parameters (like temperature, max_tokens, top_p, stop_sequences, etc.) per-request using the provider key within the options object of generateText, streamText, or generateObject.

// Example: Overriding temperature for a specific call
const response = await agent.generateText("Write a creative story.", {
  provider: {
    temperature: 0.9,
    max_tokens: 1024,
    // Any other valid Anthropic parameters
  },
});

Refer to the Anthropic API documentation for the full list of available configuration parameters.

Table of Contents