Claude 3.5 Sonnet API Tutorial

Overview

This project serves as a comprehensive tutorial on how to use the Claude 3.5 Sonnet API. It demonstrates three distinct use cases:

1. Quick Start: Basic API call
2. Tool Use: Creating and utilizing custom tools
3. Image Analysis: Loading and analyzing images

Each use case is implemented in a separate Python script, providing practical examples of interacting with the Claude 3.5 Sonnet API.

Setting Up

Environment and Requirements

1. Ensure you have Python 3.7 or later installed on your system.
2. Clone this repository to your local machine.
3. Install the required packages by running:

pip install -r requirements.txt

The requirements.txt file includes:

• anthropic: The official Anthropic Python client
• pytest: For running unit tests (used in the tool use example)

Setting up the API Key

To use the Claude 3.5 Sonnet API, you need to set up your API key:

1. Sign up for an account at https://www.anthropic.com if you haven't already.
2. Obtain your API key from the Anthropic dashboard.
3. Set the API key as an environment variable:

export ANTHROPIC_API_KEY='your_api_key_here'

Alternatively, you can set the API key in your Python script:

import anthropic
client = anthropic.Anthropic(api_key='your_api_key_here')

1. Quick Start

The 1_claude_quickstart.py script demonstrates how to make a basic API call to Claude 3.5 Sonnet.

Key points:

• Imports the anthropic library
• Creates an Anthropic client
• Sends a message to the API with specific parameters
• Prints the response

To run:

python 1_claude_quickstart.py

This script sets Claude as a world-class poet and asks it to explain why the ocean is salty in the form of a short poem.

2. Tool Use

The 2_claude_tool_use.py script showcases how to create and use custom tools with Claude 3.5 Sonnet.

Creating Tool Functions

Tool functions are defined in tools.py:

• list_artifacts(): Lists all files in the ./artifacts/ directory
• save_artifact(content, filename): Saves content to a file in the ./artifacts/ directory
• load_artifact(filename): Loads content from a file in the ./artifacts/ directory

Defining Tool Schemas

Tool schemas are JSON files in the tool_schema/ directory:

• list-artifacts.json
• save-artifact.json
• load-artifact.json

These schemas define the input and output structure for each tool.

Using Tools with Claude

The script demonstrates how to:

1. Load tool schemas
2. Create a message with tool use capabilities
3. Handle tool use responses
4. Process tool outputs

To run:

python 2_claude_tool_use.py

This script uses Claude to write a Python program that adds two numbers and creates a unit test for it, demonstrating file creation and manipulation through custom tools.

You should find the files being generated into the ./artifacts folder. You can even run the test with:

pytest

3. Image Analysis

The 3_claude_vision.py script shows how to load a local image and send it to the Claude API for analysis.

Key steps:

1. Load and encode an image file to base64
2. Create a message with both text and image content
3. Send the message to the Claude API
4. Process and display the response

To run:

python 3_claude_vision.py

This script loads an image of an AWS Lambda pricing table and asks Claude to generate a markdown table of the pricing information and model the cost for different numbers of users.