Function Calling with Claude

This guide shows how to integrate Anthropic’s Claude LLM with Desearch using Claude's Tool Use (function calling) feature. The integration lets Claude intelligently decide when to call Desearch’s AI search and return up-to-date web content.

1. Installation and Environment Setup

Make sure you have <Anchor label="python3" target="_blank" href="https://www.python.org/doc/">python3</Anchor> and <Anchor label="pip" target="_blank" href="https://pip.pypa.io/en/stable/installation/">pip</Anchor> installed on your machine.

Install the required packages:

shell

pip install anthropic desearch_py rich python-dotenv

This installs:

anthropic for Claude API calls
desearch_py for Desearch AI search
rich for better terminal output
python-dotenv for loading environment variables

API Key Setup

Create a .env file in your project root:

text

ANTHROPIC_API_KEY=your_claude_api_key_here
DESEARCH_API_KEY=your_desearch_api_key_here

Get your keys from:

<Anchor label="Anthropic Console" target="_blank" href="https://console.anthropic.com/settings/keys">Anthropic Console</Anchor>
<Anchor label="Desearch Dashboard" target="_blank" href="https://console.desearch.ai/api-keys">Desearch Dashboard</Anchor>

Make sure to add .env to .gitignore to avoid leaking secrets.

2. Understanding Tool Use in Claude

Claude can generate structured requests to call external tools you define.

Here's the tool schema you'll define for Claude:

json

{
  "name": "desearch_search",
  "description": "Perform a web search and return the most relevant data.",
  "input_schema": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "The search query to perform."
      }
    },
    "required": ["query"]
  }
}

When Claude detects the need to search, it will return something like:

json

{
  "type": "tool_use",
  "id": "toolu_xxx",
  "name": "desearch_search",
  "input": {"query": "latest news on AI regulation"}
}

You must handle this tool call and pass the query to Desearch in your own code.

3. Using Desearch with Claude Tool Use

Step 1: Initialize Clients

python

from dotenv import load_dotenv
import anthropic
import os
from desearch_py import Desearch

load_dotenv()

claude = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))
desearch = Desearch(api_key=os.getenv("DESEARCH_API_KEY"))

Step 2: Define the Tool

python

TOOLS = [
    {
        "name": "desearch_search",
        "description": "Perform a web search and return relevant data.",
        "input_schema": {
            "type": "object",
            "properties": {
                "query": {
                    "type": "string",
                    "description": "The search query to perform."
                }
            },
            "required": ["query"]
        }
    }
]

Step 3: System Prompt

python

SYSTEM_MESSAGE = "You are an assistant with access to an advanced AI search engine. Use the Desearch tool to help the user find up-to-date information."

Step 4: Define the Tool Function

python

def desearch_search(query: str) -> dict:
    return desearch.ai_search(
        prompt=query,
        tools=["twitter"],
        model="NOVA",
        date_filter="PAST_24_HOURS",
        streaming=False
    )

Step 5: Tool Call Handler

python

def process_tool_calls(tool_calls):
    search_results = []
    for tool_call in tool_calls:
        if tool_call.name == "desearch_search":
            args = tool_call.input
            results = desearch_search(**args)
            search_results.append(results)
    return search_results

4. Main Loop: Chat with Search Integration

python

from rich.console import Console
from rich.prompt import Prompt
from rich.markdown import Markdown

console = Console()

def main():
    messages = []
    while True:
        try:
            user_input = Prompt.ask("[bold yellow]What do you want to search for?[/bold yellow]")
            messages.append({"role": "user", "content": user_input})

            response = claude.messages.create(
                model="claude-3-sonnet-20240229",
                max_tokens=1024,
                system=SYSTEM_MESSAGE,
                messages=messages,
                tools=TOOLS
            )

            tool_calls = [item for item in response.content if item.type == "tool_use"]

            if tool_calls:
                results = process_tool_calls(tool_calls)
                messages.append({"role": "assistant", "content": f"Search results: {results}"})
                messages.append({"role": "user", "content": "Summarize the findings and answer the original query."})
                
                followup = claude.messages.create(
                    model="claude-3-sonnet-20240229",
                    max_tokens=1024,
                    system=SYSTEM_MESSAGE,
                    messages=messages
                )

                final_response = followup.content[0].text
                console.print(Markdown(final_response))
                messages.append({"role": "assistant", "content": final_response})

            else:
                console.print(Markdown(response.content[0].text))
                messages.append({"role": "assistant", "content": response.content[0].text})

        except Exception as e:
            console.print(f"[bold red]Error:[/bold red] {e}")

if __name__ == "__main__":
    main()

5. Run the Script

Save everything as claude_desearch.py. Then run it:

shell

python claude_desearch.py

You’ll be prompted like this:

shell

What do you want to search for?

✅ Sample Output

shell

What do you want to search for?: What is Bittensor?
Context updated with desearch_search:  What is Bittensor?
Bittensor is a decentralized machine learning network built on blockchain. It allows users to contribute AI models to a global marketplace and earn $TAO tokens. The system promotes open collaboration and aims to democratize access to AI infrastructure.

Function Calling with Claude